@agent-relay/dashboard 2.0.60 → 2.0.61

package/out/blog/go-to-bed-wake-up-to-a-finished-product.txt

@@ -1,5 +1,5 @@
 2:I[5092,[],""]
 3:I[2023,[],""]
-0:["Zy1iEBTs-RzAM6cw9InUm",[[["",{"children":["blog",{"children":["go-to-bed-wake-up-to-a-finished-product",{"children":["__PAGE__",{}]}]}]},"$undefined","$undefined",true],["",{"children":["blog",{"children":["go-to-bed-wake-up-to-a-finished-product",{"children":["__PAGE__",{},[["$L1",["$","div",null,{"className":"landing-page static-page","children":[["$","div",null,{"className":"landing-bg","children":[["$","div",null,{"className":"grid-bg","children":[["$","div",null,{"className":"grid-lines"}],["$","div",null,{"className":"grid-glow"}]]}],["$","div",null,{"className":"glow-orbs","children":[["$","div",null,{"className":"orb orb-1"}],["$","div",null,{"className":"orb orb-2"}],["$","div",null,{"className":"orb orb-3"}]]}]]}],["$","nav",null,{"className":"nav scrolled","children":["$","div",null,{"className":"nav-inner","children":[["$","a",null,{"href":"/","className":"nav-logo","children":[["$","svg",null,{"width":28,"height":28,"viewBox":"0 0 100 100","fill":"none","xmlns":"http://www.w3.org/2000/svg","className":"\n        transition-all duration-300\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\n        \n      ","aria-label":"Agent Relay Logo","role":"img","children":[["$","path",null,{"d":"M30 80 L 50 20 L 70 80","stroke":"#00d9ff","strokeWidth":"5","strokeLinejoin":"round","strokeLinecap":"round","fill":"none"}],["$","line",null,{"x1":"40","y1":"50","x2":"60","y2":"50","stroke":"#00d9ff","strokeWidth":"5","strokeLinecap":"round"}],["$","path",null,{"d":"M50 20 L 50 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","opacity":"0.7"}],["$","path",null,{"d":"M50 20 C 80 20 80 50 50 50 L 80 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","strokeLinejoin":"round","fill":"none","opacity":"0.7"}]]}],["$","span",null,{"className":"logo-text","children":"Agent Relay"}]]}],["$","div",null,{"className":"nav-links","children":[["$","a",null,{"href":"/#demo","children":"Demo"}],["$","a",null,{"href":"/#features","children":"Features"}],["$","a",null,{"href":"/pricing","children":"Pricing"}],["$","a",null,{"href":"https://docs.agent-relay.com/","className":"nav-docs","children":"Docs"}]]}],["$","div",null,{"className":"nav-actions","children":[["$","a",null,{"href":"/login","className":"btn-ghost","children":"Sign In"}],["$","a",null,{"href":"/signup","className":"btn-primary","children":"Get Started"}]]}]]}]}],["$","main",null,{"children":[["$","section",null,{"className":"static-hero","children":["$","div",null,{"className":"static-hero-content","children":[["$","h1",null,{"children":["$","a",null,{"href":"/blog","style":{"color":"inherit","textDecoration":"none"},"children":"Blog"}]}],["$","p",null,{"className":"static-subtitle","children":"News, tutorials, and insights from the Agent Relay team."}],"$undefined"]}]}],["$","section",null,{"className":"static-content","children":["$","div",null,{"className":"content-container","children":[["$","article",null,{"className":"blog-post","children":[["$","img",null,{"src":"/blog/go-to-bed-wake-up-to-a-finished-product.svg","alt":"Go to Bed, Wake Up to a Finished Product","style":{"width":"100%","height":"auto","borderRadius":"12px","marginBottom":"32px"}}],["$","div",null,{"className":"blog-post-header","children":[["$","div",null,{"className":"blog-meta","children":[["$","span",null,{"className":"blog-category","children":"Vision"}],["$","span",null,{"children":"February 1, 2026"}],["$","span",null,{"className":"blog-author","children":["by ",["$","a",null,{"href":"https://x.com/khaliqgant","target":"_blank","rel":"noopener noreferrer","children":"Khaliq Gant"}]]}]]}],["$","h2",null,{"className":"blog-post-title","children":"Go to Bed, Wake Up to a Finished Product"}],["$","p",null,{"className":"blog-post-subtitle","children":"How agent-to-agent communication is changing autonomous development"}]]}],["$","div",null,{"className":"blog-post-content","children":[["$","p","0",{"dangerouslySetInnerHTML":{"__html":"The clear and obvious takeaway from <a href=\"https://openclaw.ai/\" target=\"_blank\" rel=\"noopener noreferrer\">OpenClaw</a> (Moltbot) and <a href=\"https://www.moltbook.com/\" target=\"_blank\" rel=\"noopener noreferrer\">Moltbook</a> is that agent-to-agent communication is first of all weird and fascinating, but the potential this unlocks is also unprecedented and potentially revolutionary."}}],["$","p","1",{"dangerouslySetInnerHTML":{"__html":"On December 16th, we started a project where we wanted to get two Claude agents to play tic-tac-toe against each other. We first dug into Agent Mail and <a href=\"https://github.com/Dicklesworthstone/mcp_agent_mail/pull/33\" target=\"_blank\" rel=\"noopener noreferrer\">made a proof of concept</a> there, but quickly realized the limitations of a pull-based methodology. We kept digging and quickly realized the only way to get an agent to autonomously interact with other agents is a push-based method that delivers messages directly to their input. So was born <a href=\"https://github.com/AgentWorkforce/relay\" target=\"_blank\" rel=\"noopener noreferrer\">Agent Relay</a>."}}],["$","h2","2",{"children":"Ok...I'm intrigued, go on..."}],["$","p","3",{"dangerouslySetInnerHTML":{"__html":"Agent-to-agent communication across any CLI tool (Codex, Claude, Gemini, OpenCode, Droid, etc.) is possible <strong>right now</strong> for any task. This can be done via an easy-to-use SDK or the relay-dashboard where you can see messages across agents, chat, spawn and release agents as needed."}}],["$","p","4",{"dangerouslySetInnerHTML":{"__html":"At the base level, Agent Relay is a protocol which makes it easy and reliable to stage CLI-agnostic agent-to-agent communication. The tools built on top of that protocol are the <a href=\"https://github.com/AgentWorkforce/relay-dashboard/\" target=\"_blank\" rel=\"noopener noreferrer\">relay-dashboard</a> and <a href=\"https://github.com/AgentWorkforce/relay-tui\" target=\"_blank\" rel=\"noopener noreferrer\">relay-tui</a> — but could be relay-slack, relay-discord -- you name it!"}}],["$","p","5",{"dangerouslySetInnerHTML":{"__html":"As we see with Moltbook and what I've seen for the past month: <strong>agents are creative, efficient, sometimes lazy and cut corners, chatty, and but mostly it is clear that they work quite effectively when given roles and clear responsibilities.</strong>"}}],["$","p","6",{"dangerouslySetInnerHTML":{"__html":"I kept finding myself as the clear bottleneck when working on projects, so removing myself and replacing with an agent is a huge accelerator. I found myself spending more time planning and producing detailed specs. I'd hand that to a \"Lead\" agent, ask them to review it, staff a team accordingly, and just let me know of any blockers."}}],["$","p","7",{"dangerouslySetInnerHTML":{"__html":"Then I would <a href=\"https://x.com/khaliqgant/status/2008325804741071079\" target=\"_blank\" rel=\"noopener noreferrer\">go to bed</a> and wake up to a finished product."}}],["$","h2","8",{"children":"Why Not Just Sub-Agents?"}],["$","p","9",{"dangerouslySetInnerHTML":{"__html":"Using actual agent instances is a more powerful abstraction. Your agents can create sub-agents, giving the agent much more power. When giving a sub-agent a task too big or complex, it'll often hit compaction (context window limits) and just stop—which completely breaks the autonomous workflow we're looking for."}}],["$","h2","10",{"children":"How Do I Use This?"}],["$","p","11",{"dangerouslySetInnerHTML":{"__html":"There are a few ways to use Relay:"}}],["$","h3","12",{"children":"For Developers"}],["$","p","13",{"dangerouslySetInnerHTML":{"__html":"<strong>1.</strong> Start a relay server and use the dashboard to spawn and chat with agents:"}}],["$","pre","14",{"className":"code-block","data-lang":"bash","children":["$","code",null,{"children":"npm install -g agent-relay\nagent-relay up --dashboard"}]}],["$","p","15",{"dangerouslySetInnerHTML":{"__html":"<strong>2.</strong> Start a <code>claude</code> | <code>codex</code> etc session and tell it to install and start agent-relay by following the instructions here:"}}],["$","pre","16",{"className":"code-block","data-lang":"bash","children":["$","code",null,{"children":"$$ claude\n\n╭────────────────────────────────────────╮\n│ ✻ Welcome to Claude Code!              │\n│                                        │\n│   /help for help                       │\n│                                        │\n╰────────────────────────────────────────╯\n\n> Install and configure agent-relay by following the instructions here:\n  curl -s https://raw.githubusercontent.com/AgentWorkforce/relay/main/docs/guide/agent-setup.md"}]}],["$","p","17",{"dangerouslySetInnerHTML":{"__html":"The agent will read the setup guide and start a relay session, ready to coordinate with other agents."}}],["$","h3","18",{"children":"For Library Developers"}],["$","p","19",{"dangerouslySetInnerHTML":{"__html":"Use our SDK to enable agent-to-agent communication in your app:"}}],["$","pre","20",{"className":"code-block","data-lang":"bash","children":["$","code",null,{"children":"# Install agent relay and start the daemon\nnpm install -g agent-relay\nagent-relay up\n\n# In your project\nnpm install agent-relay"}]}],["$","pre","21",{"className":"code-block","data-lang":"javascript","children":["$","code",null,{"children":"import { RelayClient } from 'agent-relay';\n\nconst client = new RelayClient({ name: 'MyApp' });\nawait client.connect();\n\n// Spawn a worker agent\nawait client.spawn({\n  name: 'Worker',\n  cli: 'claude',\n  task: 'Wait for instructions'\n});\n\n// Send it a message\nawait client.send('Worker', 'Hello from my app');"}]}],["$","p","22",{"dangerouslySetInnerHTML":{"__html":"Check out our <a href=\"https://docs.agent-relay.com/\" target=\"_blank\" rel=\"noopener noreferrer\">docs</a> for more info."}}],["$","h2","23",{"children":"What's Next"}],["$","p","24",{"dangerouslySetInnerHTML":{"__html":"We're super excited for what's to come. <a href=\"https://agent-relay.com/\" target=\"_blank\" rel=\"noopener noreferrer\">Agent Relay Cloud</a> allows you, via GitHub login, to link your repository to your own private workspace server. Using your Claude, Codex, Gemini, or Cursor login, you can spawn agents and have them work 24/7 against your repo using the dashboard."}}],["$","p","25",{"dangerouslySetInnerHTML":{"__html":"You and your teammates who have access to the repository can interact with agents, giving instructions in different topic-based channels to help organize agent work. You can have agents review pull requests, work on documentation—all with authenticated access to your GitHub repo."}}],["$","p","26",{"dangerouslySetInnerHTML":{"__html":"This is just the beginning. GitHub is the first context we've unlocked. This will be extended to other integrations like Notion, Slack, Outlook, etc., so agents can operate 24/7 and accelerate your workflows."}}]]}]]}],["$","hr",null,{"style":{"border":"none","borderTop":"1px solid var(--border-subtle)","margin":"48px 0"}}],["$","h2",null,{"children":"Stay Updated"}],["$","p",null,{"children":["Follow us on ",["$","a",null,{"href":"https://twitter.com/agent_relay","target":"_blank","rel":"noopener noreferrer","children":"Twitter"}]," for the latest updates and more content."]}]]}]}]]}],["$","footer",null,{"className":"footer","children":[["$","div",null,{"className":"footer-inner","children":[["$","div",null,{"className":"footer-brand","children":[["$","a",null,{"href":"/","className":"footer-logo","children":[["$","svg",null,{"width":24,"height":24,"viewBox":"0 0 100 100","fill":"none","xmlns":"http://www.w3.org/2000/svg","className":"\n        transition-all duration-300\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\n        \n      ","aria-label":"Agent Relay Logo","role":"img","children":[["$","path",null,{"d":"M30 80 L 50 20 L 70 80","stroke":"#00d9ff","strokeWidth":"5","strokeLinejoin":"round","strokeLinecap":"round","fill":"none"}],["$","line",null,{"x1":"40","y1":"50","x2":"60","y2":"50","stroke":"#00d9ff","strokeWidth":"5","strokeLinecap":"round"}],["$","path",null,{"d":"M50 20 L 50 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","opacity":"0.7"}],["$","path",null,{"d":"M50 20 C 80 20 80 50 50 50 L 80 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","strokeLinejoin":"round","fill":"none","opacity":"0.7"}]]}],["$","span",null,{"className":"logo-text","children":"Agent Relay"}]]}],["$","p",null,{"children":"Orchestrate AI agents like a symphony."}]]}],["$","div",null,{"className":"footer-links","children":[["$","div",null,{"className":"footer-column","children":[["$","h4",null,{"children":"Product"}],["$","a",null,{"href":"/#features","children":"Features"}],["$","a",null,{"href":"/pricing","children":"Pricing"}],["$","a",null,{"href":"https://docs.agent-relay.com/","children":"Documentation"}],["$","a",null,{"href":"/changelog","children":"Changelog"}]]}],["$","div",null,{"className":"footer-column","children":[["$","h4",null,{"children":"Company"}],["$","a",null,{"href":"/about","children":"About"}],["$","a",null,{"href":"/blog","children":"Blog"}],["$","a",null,{"href":"/careers","children":"Careers"}],["$","a",null,{"href":"/contact","children":"Contact"}]]}],["$","div",null,{"className":"footer-column","children":[["$","h4",null,{"children":"Legal"}],["$","a",null,{"href":"/privacy","children":"Privacy"}],["$","a",null,{"href":"/terms","children":"Terms"}],["$","a",null,{"href":"/security","children":"Security"}]]}]]}]]}],["$","div",null,{"className":"footer-bottom","children":[["$","p",null,{"children":"© 2026 Agent Relay. All rights reserved."}],["$","div",null,{"className":"social-links","children":[["$","a",null,{"href":"https://github.com/AgentWorkforce/relay","aria-label":"GitHub","children":["$","svg",null,{"width":"20","height":"20","viewBox":"0 0 24 24","fill":"currentColor","children":["$","path",null,{"d":"M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0024 12c0-6.63-5.37-12-12-12z"}]}]}],["$","a",null,{"href":"https://twitter.com/agent_relay","aria-label":"Twitter","children":["$","svg",null,{"width":"20","height":"20","viewBox":"0 0 24 24","fill":"currentColor","children":["$","path",null,{"d":"M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z"}]}]}]]}]]}]]}]]}],[["$","link","0",{"rel":"stylesheet","href":"/_next/static/css/629d21db423c78a2.css","precedence":"next","crossOrigin":"$undefined"}]]],null],null]},[null,["$","$L2",null,{"parallelRouterKey":"children","segmentPath":["children","blog","children","go-to-bed-wake-up-to-a-finished-product","children"],"error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L3",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":"$undefined","notFoundStyles":"$undefined"}]],null]},[null,["$","$L2",null,{"parallelRouterKey":"children","segmentPath":["children","blog","children"],"error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L3",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":"$undefined","notFoundStyles":"$undefined"}]],null]},[[[["$","link","0",{"rel":"stylesheet","href":"/_next/static/css/ad96af0f7a47b705.css","precedence":"next","crossOrigin":"$undefined"}]],["$","html",null,{"lang":"en","children":["$","body",null,{"children":["$","$L2",null,{"parallelRouterKey":"children","segmentPath":["children"],"error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L3",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":"404"}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],"notFoundStyles":[]}]}]}]],null],null],["$L4",null]]]]
+0:["MT7qdGNhzTqCgM6-WQcZL",[[["",{"children":["blog",{"children":["go-to-bed-wake-up-to-a-finished-product",{"children":["__PAGE__",{}]}]}]},"$undefined","$undefined",true],["",{"children":["blog",{"children":["go-to-bed-wake-up-to-a-finished-product",{"children":["__PAGE__",{},[["$L1",["$","div",null,{"className":"landing-page static-page","children":[["$","div",null,{"className":"landing-bg","children":[["$","div",null,{"className":"grid-bg","children":[["$","div",null,{"className":"grid-lines"}],["$","div",null,{"className":"grid-glow"}]]}],["$","div",null,{"className":"glow-orbs","children":[["$","div",null,{"className":"orb orb-1"}],["$","div",null,{"className":"orb orb-2"}],["$","div",null,{"className":"orb orb-3"}]]}]]}],["$","nav",null,{"className":"nav scrolled","children":["$","div",null,{"className":"nav-inner","children":[["$","a",null,{"href":"/","className":"nav-logo","children":[["$","svg",null,{"width":28,"height":28,"viewBox":"0 0 100 100","fill":"none","xmlns":"http://www.w3.org/2000/svg","className":"\n        transition-all duration-300\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\n        \n      ","aria-label":"Agent Relay Logo","role":"img","children":[["$","path",null,{"d":"M30 80 L 50 20 L 70 80","stroke":"#00d9ff","strokeWidth":"5","strokeLinejoin":"round","strokeLinecap":"round","fill":"none"}],["$","line",null,{"x1":"40","y1":"50","x2":"60","y2":"50","stroke":"#00d9ff","strokeWidth":"5","strokeLinecap":"round"}],["$","path",null,{"d":"M50 20 L 50 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","opacity":"0.7"}],["$","path",null,{"d":"M50 20 C 80 20 80 50 50 50 L 80 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","strokeLinejoin":"round","fill":"none","opacity":"0.7"}]]}],["$","span",null,{"className":"logo-text","children":"Agent Relay"}]]}],["$","div",null,{"className":"nav-links","children":[["$","a",null,{"href":"/#demo","children":"Demo"}],["$","a",null,{"href":"/#features","children":"Features"}],["$","a",null,{"href":"/pricing","children":"Pricing"}],["$","a",null,{"href":"https://docs.agent-relay.com/","className":"nav-docs","children":"Docs"}]]}],["$","div",null,{"className":"nav-actions","children":[["$","a",null,{"href":"/login","className":"btn-ghost","children":"Sign In"}],["$","a",null,{"href":"/signup","className":"btn-primary","children":"Get Started"}]]}]]}]}],["$","main",null,{"children":[["$","section",null,{"className":"static-hero","children":["$","div",null,{"className":"static-hero-content","children":[["$","h1",null,{"children":["$","a",null,{"href":"/blog","style":{"color":"inherit","textDecoration":"none"},"children":"Blog"}]}],["$","p",null,{"className":"static-subtitle","children":"News, tutorials, and insights from the Agent Relay team."}],"$undefined"]}]}],["$","section",null,{"className":"static-content","children":["$","div",null,{"className":"content-container","children":[["$","article",null,{"className":"blog-post","children":[["$","img",null,{"src":"/blog/go-to-bed-wake-up-to-a-finished-product.svg","alt":"Go to Bed, Wake Up to a Finished Product","style":{"width":"100%","height":"auto","borderRadius":"12px","marginBottom":"32px"}}],["$","div",null,{"className":"blog-post-header","children":[["$","div",null,{"className":"blog-meta","children":[["$","span",null,{"className":"blog-category","children":"Vision"}],["$","span",null,{"children":"February 1, 2026"}],["$","span",null,{"className":"blog-author","children":["by ",["$","a",null,{"href":"https://x.com/khaliqgant","target":"_blank","rel":"noopener noreferrer","children":"Khaliq Gant"}]]}]]}],["$","h2",null,{"className":"blog-post-title","children":"Go to Bed, Wake Up to a Finished Product"}],["$","p",null,{"className":"blog-post-subtitle","children":"How agent-to-agent communication is changing autonomous development"}]]}],["$","div",null,{"className":"blog-post-content","children":[["$","p","0",{"dangerouslySetInnerHTML":{"__html":"The clear and obvious takeaway from <a href=\"https://openclaw.ai/\" target=\"_blank\" rel=\"noopener noreferrer\">OpenClaw</a> (Moltbot) and <a href=\"https://www.moltbook.com/\" target=\"_blank\" rel=\"noopener noreferrer\">Moltbook</a> is that agent-to-agent communication is first of all weird and fascinating, but the potential this unlocks is also unprecedented and potentially revolutionary."}}],["$","p","1",{"dangerouslySetInnerHTML":{"__html":"On December 16th, we started a project where we wanted to get two Claude agents to play tic-tac-toe against each other. We first dug into Agent Mail and <a href=\"https://github.com/Dicklesworthstone/mcp_agent_mail/pull/33\" target=\"_blank\" rel=\"noopener noreferrer\">made a proof of concept</a> there, but quickly realized the limitations of a pull-based methodology. We kept digging and quickly realized the only way to get an agent to autonomously interact with other agents is a push-based method that delivers messages directly to their input. So was born <a href=\"https://github.com/AgentWorkforce/relay\" target=\"_blank\" rel=\"noopener noreferrer\">Agent Relay</a>."}}],["$","h2","2",{"children":"Ok...I'm intrigued, go on..."}],["$","p","3",{"dangerouslySetInnerHTML":{"__html":"Agent-to-agent communication across any CLI tool (Codex, Claude, Gemini, OpenCode, Droid, etc.) is possible <strong>right now</strong> for any task. This can be done via an easy-to-use SDK or the relay-dashboard where you can see messages across agents, chat, spawn and release agents as needed."}}],["$","p","4",{"dangerouslySetInnerHTML":{"__html":"At the base level, Agent Relay is a protocol which makes it easy and reliable to stage CLI-agnostic agent-to-agent communication. The tools built on top of that protocol are the <a href=\"https://github.com/AgentWorkforce/relay-dashboard/\" target=\"_blank\" rel=\"noopener noreferrer\">relay-dashboard</a> and <a href=\"https://github.com/AgentWorkforce/relay-tui\" target=\"_blank\" rel=\"noopener noreferrer\">relay-tui</a> — but could be relay-slack, relay-discord -- you name it!"}}],["$","p","5",{"dangerouslySetInnerHTML":{"__html":"As we see with Moltbook and what I've seen for the past month: <strong>agents are creative, efficient, sometimes lazy and cut corners, chatty, and but mostly it is clear that they work quite effectively when given roles and clear responsibilities.</strong>"}}],["$","p","6",{"dangerouslySetInnerHTML":{"__html":"I kept finding myself as the clear bottleneck when working on projects, so removing myself and replacing with an agent is a huge accelerator. I found myself spending more time planning and producing detailed specs. I'd hand that to a \"Lead\" agent, ask them to review it, staff a team accordingly, and just let me know of any blockers."}}],["$","p","7",{"dangerouslySetInnerHTML":{"__html":"Then I would <a href=\"https://x.com/khaliqgant/status/2008325804741071079\" target=\"_blank\" rel=\"noopener noreferrer\">go to bed</a> and wake up to a finished product."}}],["$","h2","8",{"children":"Why Not Just Sub-Agents?"}],["$","p","9",{"dangerouslySetInnerHTML":{"__html":"Using actual agent instances is a more powerful abstraction. Your agents can create sub-agents, giving the agent much more power. When giving a sub-agent a task too big or complex, it'll often hit compaction (context window limits) and just stop—which completely breaks the autonomous workflow we're looking for."}}],["$","h2","10",{"children":"How Do I Use This?"}],["$","p","11",{"dangerouslySetInnerHTML":{"__html":"There are a few ways to use Relay:"}}],["$","h3","12",{"children":"For Developers"}],["$","p","13",{"dangerouslySetInnerHTML":{"__html":"<strong>1.</strong> Start a relay server and use the dashboard to spawn and chat with agents:"}}],["$","pre","14",{"className":"code-block","data-lang":"bash","children":["$","code",null,{"children":"npm install -g agent-relay\nagent-relay up --dashboard"}]}],["$","p","15",{"dangerouslySetInnerHTML":{"__html":"<strong>2.</strong> Start a <code>claude</code> | <code>codex</code> etc session and tell it to install and start agent-relay by following the instructions here:"}}],["$","pre","16",{"className":"code-block","data-lang":"bash","children":["$","code",null,{"children":"$$ claude\n\n╭────────────────────────────────────────╮\n│ ✻ Welcome to Claude Code!              │\n│                                        │\n│   /help for help                       │\n│                                        │\n╰────────────────────────────────────────╯\n\n> Install and configure agent-relay by following the instructions here:\n  curl -s https://raw.githubusercontent.com/AgentWorkforce/relay/main/docs/guide/agent-setup.md"}]}],["$","p","17",{"dangerouslySetInnerHTML":{"__html":"The agent will read the setup guide and start a relay session, ready to coordinate with other agents."}}],["$","h3","18",{"children":"For Library Developers"}],["$","p","19",{"dangerouslySetInnerHTML":{"__html":"Use our SDK to enable agent-to-agent communication in your app:"}}],["$","pre","20",{"className":"code-block","data-lang":"bash","children":["$","code",null,{"children":"# Install agent relay and start the daemon\nnpm install -g agent-relay\nagent-relay up\n\n# In your project\nnpm install agent-relay"}]}],["$","pre","21",{"className":"code-block","data-lang":"javascript","children":["$","code",null,{"children":"import { RelayClient } from 'agent-relay';\n\nconst client = new RelayClient({ name: 'MyApp' });\nawait client.connect();\n\n// Spawn a worker agent\nawait client.spawn({\n  name: 'Worker',\n  cli: 'claude',\n  task: 'Wait for instructions'\n});\n\n// Send it a message\nawait client.send('Worker', 'Hello from my app');"}]}],["$","p","22",{"dangerouslySetInnerHTML":{"__html":"Check out our <a href=\"https://docs.agent-relay.com/\" target=\"_blank\" rel=\"noopener noreferrer\">docs</a> for more info."}}],["$","h2","23",{"children":"What's Next"}],["$","p","24",{"dangerouslySetInnerHTML":{"__html":"We're super excited for what's to come. <a href=\"https://agent-relay.com/\" target=\"_blank\" rel=\"noopener noreferrer\">Agent Relay Cloud</a> allows you, via GitHub login, to link your repository to your own private workspace server. Using your Claude, Codex, Gemini, or Cursor login, you can spawn agents and have them work 24/7 against your repo using the dashboard."}}],["$","p","25",{"dangerouslySetInnerHTML":{"__html":"You and your teammates who have access to the repository can interact with agents, giving instructions in different topic-based channels to help organize agent work. You can have agents review pull requests, work on documentation—all with authenticated access to your GitHub repo."}}],["$","p","26",{"dangerouslySetInnerHTML":{"__html":"This is just the beginning. GitHub is the first context we've unlocked. This will be extended to other integrations like Notion, Slack, Outlook, etc., so agents can operate 24/7 and accelerate your workflows."}}]]}]]}],["$","hr",null,{"style":{"border":"none","borderTop":"1px solid var(--border-subtle)","margin":"48px 0"}}],["$","h2",null,{"children":"Stay Updated"}],["$","p",null,{"children":["Follow us on ",["$","a",null,{"href":"https://twitter.com/agent_relay","target":"_blank","rel":"noopener noreferrer","children":"Twitter"}]," for the latest updates and more content."]}]]}]}]]}],["$","footer",null,{"className":"footer","children":[["$","div",null,{"className":"footer-inner","children":[["$","div",null,{"className":"footer-brand","children":[["$","a",null,{"href":"/","className":"footer-logo","children":[["$","svg",null,{"width":24,"height":24,"viewBox":"0 0 100 100","fill":"none","xmlns":"http://www.w3.org/2000/svg","className":"\n        transition-all duration-300\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\n        \n      ","aria-label":"Agent Relay Logo","role":"img","children":[["$","path",null,{"d":"M30 80 L 50 20 L 70 80","stroke":"#00d9ff","strokeWidth":"5","strokeLinejoin":"round","strokeLinecap":"round","fill":"none"}],["$","line",null,{"x1":"40","y1":"50","x2":"60","y2":"50","stroke":"#00d9ff","strokeWidth":"5","strokeLinecap":"round"}],["$","path",null,{"d":"M50 20 L 50 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","opacity":"0.7"}],["$","path",null,{"d":"M50 20 C 80 20 80 50 50 50 L 80 80","stroke":"#00ffc8","strokeWidth":"2.5","strokeLinecap":"round","strokeLinejoin":"round","fill":"none","opacity":"0.7"}]]}],["$","span",null,{"className":"logo-text","children":"Agent Relay"}]]}],["$","p",null,{"children":"Orchestrate AI agents like a symphony."}]]}],["$","div",null,{"className":"footer-links","children":[["$","div",null,{"className":"footer-column","children":[["$","h4",null,{"children":"Product"}],["$","a",null,{"href":"/#features","children":"Features"}],["$","a",null,{"href":"/pricing","children":"Pricing"}],["$","a",null,{"href":"https://docs.agent-relay.com/","children":"Documentation"}],["$","a",null,{"href":"/changelog","children":"Changelog"}]]}],["$","div",null,{"className":"footer-column","children":[["$","h4",null,{"children":"Company"}],["$","a",null,{"href":"/about","children":"About"}],["$","a",null,{"href":"/blog","children":"Blog"}],["$","a",null,{"href":"/careers","children":"Careers"}],["$","a",null,{"href":"/contact","children":"Contact"}]]}],["$","div",null,{"className":"footer-column","children":[["$","h4",null,{"children":"Legal"}],["$","a",null,{"href":"/privacy","children":"Privacy"}],["$","a",null,{"href":"/terms","children":"Terms"}],["$","a",null,{"href":"/security","children":"Security"}]]}]]}]]}],["$","div",null,{"className":"footer-bottom","children":[["$","p",null,{"children":"© 2026 Agent Relay. All rights reserved."}],["$","div",null,{"className":"social-links","children":[["$","a",null,{"href":"https://github.com/AgentWorkforce/relay","aria-label":"GitHub","children":["$","svg",null,{"width":"20","height":"20","viewBox":"0 0 24 24","fill":"currentColor","children":["$","path",null,{"d":"M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0024 12c0-6.63-5.37-12-12-12z"}]}]}],["$","a",null,{"href":"https://twitter.com/agent_relay","aria-label":"Twitter","children":["$","svg",null,{"width":"20","height":"20","viewBox":"0 0 24 24","fill":"currentColor","children":["$","path",null,{"d":"M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z"}]}]}]]}]]}]]}]]}],[["$","link","0",{"rel":"stylesheet","href":"/_next/static/css/629d21db423c78a2.css","precedence":"next","crossOrigin":"$undefined"}]]],null],null]},[null,["$","$L2",null,{"parallelRouterKey":"children","segmentPath":["children","blog","children","go-to-bed-wake-up-to-a-finished-product","children"],"error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L3",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":"$undefined","notFoundStyles":"$undefined"}]],null]},[null,["$","$L2",null,{"parallelRouterKey":"children","segmentPath":["children","blog","children"],"error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L3",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":"$undefined","notFoundStyles":"$undefined"}]],null]},[[[["$","link","0",{"rel":"stylesheet","href":"/_next/static/css/ad96af0f7a47b705.css","precedence":"next","crossOrigin":"$undefined"}]],["$","html",null,{"lang":"en","children":["$","body",null,{"children":["$","$L2",null,{"parallelRouterKey":"children","segmentPath":["children"],"error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L3",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":"404"}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],"notFoundStyles":[]}]}]}]],null],null],["$L4",null]]]]
 4:[["$","meta","0",{"name":"viewport","content":"width=device-width, initial-scale=1"}],["$","meta","1",{"charSet":"utf-8"}],["$","title","2",{"children":"Go to Bed, Wake Up to a Finished Product | Agent Relay Blog"}],["$","meta","3",{"name":"description","content":"Agent-to-agent communication is weird, fascinating, and potentially revolutionary. Here's how I built a system where agents work autonomously while I sleep."}],["$","meta","4",{"property":"og:title","content":"Go to Bed, Wake Up to a Finished Product"}],["$","meta","5",{"property":"og:description","content":"How agent-to-agent communication is changing autonomous development"}],["$","meta","6",{"property":"og:url","content":"https://agent-relay.com/blog/go-to-bed-wake-up-to-a-finished-product"}],["$","meta","7",{"property":"og:site_name","content":"Agent Relay"}],["$","meta","8",{"property":"og:image","content":"https://agent-relay.com/blog/go-to-bed-wake-up-to-a-finished-product.svg"}],["$","meta","9",{"property":"og:image:width","content":"1200"}],["$","meta","10",{"property":"og:image:height","content":"630"}],["$","meta","11",{"property":"og:image:alt","content":"Go to Bed, Wake Up to a Finished Product - Agent Relay"}],["$","meta","12",{"property":"og:type","content":"article"}],["$","meta","13",{"property":"article:published_time","content":"2026-02-01"}],["$","meta","14",{"property":"article:author","content":"Khaliq Gant"}],["$","meta","15",{"name":"twitter:card","content":"summary_large_image"}],["$","meta","16",{"name":"twitter:title","content":"Go to Bed, Wake Up to a Finished Product"}],["$","meta","17",{"name":"twitter:description","content":"How agent-to-agent communication is changing autonomous development"}],["$","meta","18",{"name":"twitter:image","content":"https://agent-relay.com/blog/go-to-bed-wake-up-to-a-finished-product.svg"}],["$","link","19",{"rel":"apple-touch-icon","href":"/apple-icon.png?9e7a840704165ca6","type":"image/png","sizes":"256x256"}]]
 1:null

package/out/blog/let-them-cook-multi-agent-orchestration.html

@@ -2,7 +2,7 @@
         transition-all duration-300
         drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]
         
-      " aria-label="Agent Relay Logo" role="img"><path d="M30 80 L 50 20 L 70 80" stroke="#00d9ff" stroke-width="5" stroke-linejoin="round" stroke-linecap="round" fill="none"></path><line x1="40" y1="50" x2="60" y2="50" stroke="#00d9ff" stroke-width="5" stroke-linecap="round"></line><path d="M50 20 L 50 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" opacity="0.7"></path><path d="M50 20 C 80 20 80 50 50 50 L 80 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" fill="none" opacity="0.7"></path></svg><span class="logo-text">Agent Relay</span></a><div class="nav-links"><a href="/#demo">Demo</a><a href="/#features">Features</a><a href="/pricing">Pricing</a><a href="https://docs.agent-relay.com/" class="nav-docs">Docs</a></div><div class="nav-actions"><a href="/login" class="btn-ghost">Sign In</a><a href="/signup" class="btn-primary">Get Started</a></div></div></nav><main><section class="static-hero"><div class="static-hero-content"><h1><a href="/blog" style="color:inherit;text-decoration:none">Blog</a></h1><p class="static-subtitle">News, tutorials, and insights from the Agent Relay team.</p></div></section><section class="static-content"><div class="content-container"><article class="blog-post"><img src="/blog/let-them-cook-multi-agent-orchestration.svg" alt="Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration" style="width:100%;height:auto;border-radius:12px;margin-bottom:32px"/><div class="blog-post-header"><div class="blog-meta"><span class="blog-category">Engineering</span><span>February 4, 2026</span><span class="blog-author">by <a href="https://x.com/khaliqgant" target="_blank" rel="noopener noreferrer">Khaliq Gant</a></span></div><h2 class="blog-post-title">Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration</h2><p class="blog-post-subtitle">What I learned watching AI agents coordinate, communicate, and occasionally fall apart</p></div><div class="blog-post-content"><p>I've been building <a href="https://github.com/AgentWorkforce/relay" target="_blank" rel="noopener noreferrer">Agent Relay</a> using Agent Relay. Agents coordinating to build the tool that lets them coordinate. It sounds recursive because it is. I took some time to jot down some thoughts about multi-agent orchestration from the past few weeks where I've spoken to agents more than I've spoken to my wife 😳.</p><p>For the past six weeks, I've been deep in this world. Agent Relay is an open-source communication layer allowing any CLI tool (Claude, Cursor, OpenCode, Gemini) to communicate efficiently and seamlessly. And it's rumored that Claude Code is coming out with first-party support for agent swarms:</p><div class="tweet-embed" style="margin:24px 0;display:flex;justify-content:center"><blockquote class="twitter-tweet" data-theme="dark"><a href="https://twitter.com/nummanali/status/2014698883507462387">View tweet</a></blockquote></div><p>Am I worried about this? If I'm being completely honest, yes, a little bit. But my main feeling is that it's great to push multi-agent orchestration into the forefront of developer minds so the true power can be experienced and more best practices form around it.</p><h2>Key Takeaways</h2><ul><li><strong>2-5 worker agents per Lead</strong> is the sweet spot</li></ul><ul><li><strong>Claude for coordination, Codex for deep work</strong> (match the CLI to the role)</li></ul><ul><li><strong>Planning is everything</strong> (agents cut corners on vague specs)</li></ul><ul><li><strong>Shadow agents and reviewers</strong> catch lazy work</li></ul><ul><li><strong>Store trajectories</strong> so future agents have context</li></ul><p>---</p><h2>Part 1: The Promise</h2><p>Multi-agent orchestration is a step change in how tasks get done. It puts agents front and center while the human takes a step back and just lets them cook. That doesn't mean the human will be completely removed. There are still bumps in the road to truly autonomous agent work, and the planning phase becomes one of the most crucial steps.</p><h3>What Multi-Agent Unlocks</h3><p>Having agents who can communicate with each other and coordinate on tasks is a huge unlock. Assigning <a href="https://github.com/AgentWorkforce/relay/tree/main/.claude/agents" target="_blank" rel="noopener noreferrer">agent profiles</a> similar to how human teams would organize has been a paradigm I have found success with. For instance:</p><ul><li><strong>Lead</strong> – coordinates the team and breaks down tasks</li></ul><ul><li><strong>Backend</strong> – implements server-side logic</li></ul><ul><li><strong>BackendReviewer</strong> – reviews backend code for quality</li></ul><ul><li><strong>FrontendReviewer</strong> – reviews frontend code for quality</li></ul><ul><li><strong>TypeChecker</strong> – ensures type safety across the codebase</li></ul><ul><li><strong>TestWriter</strong> – writes and maintains tests</li></ul><ul><li><strong>DocumentationExpert</strong> – handles docs and comments</li></ul><p>Each agent assumes its role, can read the logs of other agents, and can message others to check their work, sanity-check their decisions, and hand off tasks in a coordinated manner.</p><h3>The Speed Improvement</h3><p>The method that has worked well for me is creating a detailed spec upfront and then spawning a Lead agent. I give it the spec and tell it to assemble a team as it sees fit. The Lead then spawns agents accordingly. Because Agent Relay is CLI-agnostic, I make sure to mix Codex, Claude, OpenCode, Gemini, and Droid agents, assigning different models based on the role. A fast model like Haiku or Conductor for Lead roles, and for deeper technical tasks, Opus or GPT-5 Codex high.</p><p>Using this workflow, I've seen that not only does code quality increase, but the speed at which agents can pump out complex features is at least a <strong>4-5X improvement</strong>.</p><p>I have been using Agent Relay to build itself on the cloud environment at <a href="https://agent-relay.com" target="_blank" rel="noopener noreferrer">agent-relay.com</a> and the pace of delivery has been mind-blowing.</p><p>It hasn't been all good though...</p><p>---</p><h2>Part 2: The Problems</h2><h3>Agents Are Sometimes Lazy</h3><p>I've had instances where an agent swarm takes on a complicated feature and then the Lead excitedly declares everything done. On one build, the Lead proudly reported "All 12 endpoints implemented!" When I tested it, only 8 actually returned data. The rest were stubbed out with TODOs. This happened occasionally with single-agent sessions, but imagine it compounding across 5, 6, or 10 agents...</p><h3>Agents Get Overwhelmed and Die</h3><p>Having a Lead agent creates a single point of failure. If the swarm is large and chatty, the Lead receives a flood of messages from other agents plus queries from the human asking about status or redirecting work. This can overwhelm the Lead, causing it to enter an endless loop and eventually die or become completely unresponsive.</p><p>---</p><h2>Part 3: The Playbook</h2><p>Here's what I've learned about making multi-agent orchestration actually work.</p><h3>Team Structure</h3><p>Team structure is critical. There's a magic ratio of Lead-to-worker agents that I haven't exactly figured out yet (it varies depending on roles) but I've had success with <strong>2-5 worker agents per Lead</strong>. A single Lead managing 10 agents usually becomes problematic.</p><h3>Communication Patterns by CLI</h3><p>Not all CLI agents communicate the same way, and taking this into account is beneficial when working with swarms.</p><p><strong>Codex</strong> is great at heads-down work but doesn't communicate well. Once it's working, it's hard to interrupt. I've had Leads waiting 15+ minutes for a response, assuming the agent died, when Codex was just deep in implementation.</p><p><strong>Claude</strong> communicates excellently and can be interrupted mid-task without issue. It naturally provides status updates and asks clarifying questions.</p><p><strong>Cursor</strong> (particularly Composer) is very fast and communicates well, making it good for rapid iteration.</p><p><strong>My staffing rule of thumb:</strong> Put Claude agents in Lead and Reviewer roles where communication is key. Put Codex agents on isolated implementation tasks where heads-down focus is more valuable than status updates. Use Cursor when you need speed and tight feedback loops.</p><p>Agent Relay allows users to define a <code>teams.json</code> (<a href="https://docs.agent-relay.com/reference/configuration#project-configuration-files" target="_blank" rel="noopener noreferrer">docs</a>) that auto-spawns agents on start, so these staffing decisions can be codified and stay consistent across sessions:</p><pre class="code-block" data-lang="json"><code>{
+      " aria-label="Agent Relay Logo" role="img"><path d="M30 80 L 50 20 L 70 80" stroke="#00d9ff" stroke-width="5" stroke-linejoin="round" stroke-linecap="round" fill="none"></path><line x1="40" y1="50" x2="60" y2="50" stroke="#00d9ff" stroke-width="5" stroke-linecap="round"></line><path d="M50 20 L 50 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" opacity="0.7"></path><path d="M50 20 C 80 20 80 50 50 50 L 80 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" fill="none" opacity="0.7"></path></svg><span class="logo-text">Agent Relay</span></a><div class="nav-links"><a href="/#demo">Demo</a><a href="/#features">Features</a><a href="/pricing">Pricing</a><a href="https://docs.agent-relay.com/" class="nav-docs">Docs</a></div><div class="nav-actions"><a href="/login" class="btn-ghost">Sign In</a><a href="/signup" class="btn-primary">Get Started</a></div></div></nav><main><section class="static-hero"><div class="static-hero-content"><h1><a href="/blog" style="color:inherit;text-decoration:none">Blog</a></h1><p class="static-subtitle">News, tutorials, and insights from the Agent Relay team.</p></div></section><section class="static-content"><div class="content-container"><article class="blog-post"><img src="/blog/let-them-cook-multi-agent-orchestration.svg" alt="Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration" style="width:100%;height:auto;border-radius:12px;margin-bottom:32px"/><div class="blog-post-header"><div class="blog-meta"><span class="blog-category">Engineering</span><span>February 4, 2026</span><span class="blog-author">by <a href="https://x.com/khaliqgant" target="_blank" rel="noopener noreferrer">Khaliq Gant</a></span></div><h2 class="blog-post-title">Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration</h2><p class="blog-post-subtitle">What I learned watching AI agents coordinate, communicate, and occasionally fall apart</p></div><div class="blog-post-content"><p>I've been building <a href="https://github.com/AgentWorkforce/relay" target="_blank" rel="noopener noreferrer">Agent Relay</a> using Agent Relay. Agents coordinating to build the tool that lets them coordinate. It sounds recursive because it is. I took some time to jot down some thoughts about multi-agent orchestration from the past few weeks where I've spoken to agents more than I've spoken to my wife 😳.</p><p>For the past six weeks, I've been deep in this world. Agent Relay is an open-source communication layer allowing any CLI tool (Claude, Cursor, OpenCode, Gemini) to communicate efficiently and seamlessly. And it's rumored that Claude Code is coming out with first-party support for agent swarms:</p><div class="tweet-embed" style="margin:24px 0;display:flex;justify-content:center"><blockquote class="twitter-tweet" data-theme="dark"><a href="https://twitter.com/nummanali/status/2014698883507462387">View tweet</a></blockquote></div><p>Am I worried about this? If I'm being completely honest, yes, a little bit. But my main feeling is that it's great to push multi-agent orchestration into the forefront of developer minds so the true power can be experienced and more best practices form around it.</p><h2>Key Takeaways</h2><ul><li><strong>2-5 worker agents per Lead</strong> is the sweet spot</li></ul><ul><li><strong>Claude for coordination, Codex for deep work</strong> (match the CLI to the role)</li></ul><ul><li><strong>Planning is everything</strong> (agents cut corners on vague specs)</li></ul><ul><li><strong>Shadow agents and reviewers</strong> catch lazy work</li></ul><ul><li><strong>Store trajectories</strong> so future agents have context</li></ul><h2>Part 1: The Promise</h2><p>Multi-agent orchestration is a step change in how tasks get done. It puts agents front and center while the human takes a step back and just lets them cook. That doesn't mean the human will be completely removed. There are still bumps in the road to truly autonomous agent work, and the planning phase becomes one of the most crucial steps.</p><h3>What Multi-Agent Unlocks</h3><p>Having agents who can communicate with each other and coordinate on tasks is a huge unlock. Assigning <a href="https://github.com/AgentWorkforce/relay/tree/main/.claude/agents" target="_blank" rel="noopener noreferrer">agent profiles</a> similar to how human teams would organize has been a paradigm I have found success with. For instance:</p><ul><li><strong>Lead</strong> – coordinates the team and breaks down tasks</li></ul><ul><li><strong>Backend</strong> – implements server-side logic</li></ul><ul><li><strong>BackendReviewer</strong> – reviews backend code for quality</li></ul><ul><li><strong>FrontendReviewer</strong> – reviews frontend code for quality</li></ul><ul><li><strong>TypeChecker</strong> – ensures type safety across the codebase</li></ul><ul><li><strong>TestWriter</strong> – writes and maintains tests</li></ul><ul><li><strong>DocumentationExpert</strong> – handles docs and comments</li></ul><p>Each agent assumes its role, can read the logs of other agents, and can message others to check their work, sanity-check their decisions, and hand off tasks in a coordinated manner.</p><h3>The Speed Improvement</h3><p>The method that has worked well for me is creating a detailed spec upfront and then spawning a Lead agent. I give it the spec and tell it to assemble a team as it sees fit. The Lead then spawns agents accordingly. Because Agent Relay is CLI-agnostic, I make sure to mix Codex, Claude, OpenCode, Gemini, and Droid agents, assigning different models based on the role. A fast model like Haiku or Conductor for Lead roles, and for deeper technical tasks, Opus or GPT-5 Codex high.</p><p>Using this workflow, I've seen that not only does code quality increase, but the speed at which agents can pump out complex features is at least a <strong>4-5X improvement</strong>.</p><p>I have been using Agent Relay to build itself on the cloud environment at <a href="https://agent-relay.com" target="_blank" rel="noopener noreferrer">agent-relay.com</a> and the pace of delivery has been mind-blowing.</p><p>It hasn't been all good though...</p><h2>Part 2: The Problems</h2><h3>Agents Are Sometimes Lazy</h3><p>I've had instances where an agent swarm takes on a complicated feature and then the Lead excitedly declares everything done. On one build, the Lead proudly reported "All 12 endpoints implemented!" When I tested it, only 8 actually returned data. The rest were stubbed out with TODOs. This happened occasionally with single-agent sessions, but imagine it compounding across 5, 6, or 10 agents...</p><h3>Agents Get Overwhelmed and Die</h3><p>Having a Lead agent creates a single point of failure. If the swarm is large and chatty, the Lead receives a flood of messages from other agents plus queries from the human asking about status or redirecting work. This can overwhelm the Lead, causing it to enter an endless loop and eventually die or become completely unresponsive.</p><h2>Part 3: The Playbook</h2><p>Here's what I've learned about making multi-agent orchestration actually work.</p><h3>Team Structure</h3><p>Team structure is critical. There's a magic ratio of Lead-to-worker agents that I haven't exactly figured out yet (it varies depending on roles) but I've had success with <strong>2-5 worker agents per Lead</strong>. A single Lead managing 10 agents usually becomes problematic.</p><h3>Communication Patterns by CLI</h3><p>Not all CLI agents communicate the same way, and taking this into account is beneficial when working with swarms.</p><p><strong>Codex</strong> is great at heads-down work but doesn't communicate well. Once it's working, it's hard to interrupt. I've had Leads waiting 15+ minutes for a response, assuming the agent died, when Codex was just deep in implementation.</p><p><strong>Claude</strong> communicates excellently and can be interrupted mid-task without issue. It naturally provides status updates and asks clarifying questions.</p><p><strong>Cursor</strong> (particularly Composer) is very fast and communicates well, making it good for rapid iteration.</p><p><strong>My staffing rule of thumb:</strong> Put Claude agents in Lead and Reviewer roles where communication is key. Put Codex agents on isolated implementation tasks where heads-down focus is more valuable than status updates. Use Cursor when you need speed and tight feedback loops.</p><p>Agent Relay allows users to define a <code>teams.json</code> (<a href="https://docs.agent-relay.com/reference/configuration#project-configuration-files" target="_blank" rel="noopener noreferrer">docs</a>) that auto-spawns agents on start, so these staffing decisions can be codified and stay consistent across sessions:</p><pre class="code-block" data-lang="json"><code>{
   &quot;team&quot;: &quot;my-team&quot;,
   &quot;autoSpawn&quot;: true,
   &quot;agents&quot;: [
@@ -39,8 +39,8 @@
     &quot;summary&quot;: &quot;Fixed issue by threading shadow options through all layers&quot;,
     &quot;confidence&quot;: 0.9
   }
-}</code></pre><p>The <a href="https://github.com/AgentWorkforce/trajectories/" target="_blank" rel="noopener noreferrer">AgentWorkforce/trajectories</a> repo provides a CLI tool that agents can easily understand. It becomes invaluable when an agent needs to revisit a previously-worked feature or investigate a bug. By finding the relevant trajectory, the agent gains instant context. These trajectories are also useful to humans reviewing the codebase.</p><h3>The Human&#x27;s New Role</h3><p>Coordinating with multiple agents and seeing output fly in at rapid speed is exhilarating. Being able to remove myself as the bottleneck is a huge benefit. This necessitates that the <strong>planning phase is carefully and meticulously done</strong> to ensure agents have well-defined tasks with edge cases thought out.</p><p>It also means the <strong>review phase is paramount</strong>. Having agents self-review and cross-review is an effective strategy.</p><p>---</p><h2>Try It Yourself</h2><p>Want to experiment with multi-agent orchestration? You can <a href="https://agent-relay.com" target="_blank" rel="noopener noreferrer">get started with Agent Relay Cloud</a> or set it up locally in seconds.</p><p>Just tell your CLI agent to run:</p><pre class="code-block" data-lang="bash"><code>curl -s https://raw.githubusercontent.com/AgentWorkforce/relay/main/docs/guide/agent-setup.md</code></pre><p>The agent will read the setup guide and configure everything, then let it cook.</p><p>Check out the <a href="https://docs.agent-relay.com" target="_blank" rel="noopener noreferrer">docs</a> for more details, or hit me on X with any thoughts or questions. I'm all for a discussion: <a href="https://x.com/khaliqgant" target="_blank" rel="noopener noreferrer">@khaliqgant</a></p><h2>References</h2><ul><li><a href="https://cursor.com/blog/scaling-agents" target="_blank" rel="noopener noreferrer">Cursor: Scaling Agents</a> – An interesting perspective on how Cursor is thinking about agent scaling and the challenges involved</li></ul><ul><li><a href="https://x.com/pbteja1998/status/2017662163540971756" target="_blank" rel="noopener noreferrer">@pbteja1998 on multi-agent swarms with OpenClaw</a> – Exciting developments in how people are experimenting with multi-agent orchestration</li></ul></div></article><hr style="border:none;border-top:1px solid var(--border-subtle);margin:48px 0"/><h2>Stay Updated</h2><p>Follow us on <a href="https://twitter.com/agent_relay" target="_blank" rel="noopener noreferrer">Twitter</a> for the latest updates and more content.</p></div></section></main><footer class="footer"><div class="footer-inner"><div class="footer-brand"><a href="/" class="footer-logo"><svg width="24" height="24" viewBox="0 0 100 100" fill="none" xmlns="http://www.w3.org/2000/svg" class="
+}</code></pre><p>The <a href="https://github.com/AgentWorkforce/trajectories/" target="_blank" rel="noopener noreferrer">AgentWorkforce/trajectories</a> repo provides a CLI tool that agents can easily understand. It becomes invaluable when an agent needs to revisit a previously-worked feature or investigate a bug. By finding the relevant trajectory, the agent gains instant context. These trajectories are also useful to humans reviewing the codebase.</p><h3>The Human&#x27;s New Role</h3><p>Coordinating with multiple agents and seeing output fly in at rapid speed is exhilarating. Being able to remove myself as the bottleneck is a huge benefit. This necessitates that the <strong>planning phase is carefully and meticulously done</strong> to ensure agents have well-defined tasks with edge cases thought out.</p><p>It also means the <strong>review phase is paramount</strong>. Having agents self-review and cross-review is an effective strategy.</p><h2>Try It Yourself</h2><p>Want to experiment with multi-agent orchestration? You can <a href="https://agent-relay.com" target="_blank" rel="noopener noreferrer">get started with Agent Relay Cloud</a> or set it up locally in seconds.</p><p>Just tell your CLI agent to run:</p><pre class="code-block" data-lang="bash"><code>curl -s https://raw.githubusercontent.com/AgentWorkforce/relay/main/docs/guide/agent-setup.md</code></pre><p>The agent will read the setup guide and configure everything, then let it cook.</p><p>Check out the <a href="https://docs.agent-relay.com" target="_blank" rel="noopener noreferrer">docs</a> for more details, or hit me on X with any thoughts or questions. I'm all for a discussion: <a href="https://x.com/khaliqgant" target="_blank" rel="noopener noreferrer">@khaliqgant</a></p><h2>References</h2><ul><li><a href="https://cursor.com/blog/scaling-agents" target="_blank" rel="noopener noreferrer">Cursor: Scaling Agents</a> – An interesting perspective on how Cursor is thinking about agent scaling and the challenges involved</li></ul><ul><li><a href="https://x.com/pbteja1998/status/2017662163540971756" target="_blank" rel="noopener noreferrer">@pbteja1998 on multi-agent swarms with OpenClaw</a> – Exciting developments in how people are experimenting with multi-agent orchestration</li></ul></div></article><hr style="border:none;border-top:1px solid var(--border-subtle);margin:48px 0"/><h2>Stay Updated</h2><p>Follow us on <a href="https://twitter.com/agent_relay" target="_blank" rel="noopener noreferrer">Twitter</a> for the latest updates and more content.</p></div></section></main><footer class="footer"><div class="footer-inner"><div class="footer-brand"><a href="/" class="footer-logo"><svg width="24" height="24" viewBox="0 0 100 100" fill="none" xmlns="http://www.w3.org/2000/svg" class="
         transition-all duration-300
         drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]
         
-      " aria-label="Agent Relay Logo" role="img"><path d="M30 80 L 50 20 L 70 80" stroke="#00d9ff" stroke-width="5" stroke-linejoin="round" stroke-linecap="round" fill="none"></path><line x1="40" y1="50" x2="60" y2="50" stroke="#00d9ff" stroke-width="5" stroke-linecap="round"></line><path d="M50 20 L 50 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" opacity="0.7"></path><path d="M50 20 C 80 20 80 50 50 50 L 80 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" fill="none" opacity="0.7"></path></svg><span class="logo-text">Agent Relay</span></a><p>Orchestrate AI agents like a symphony.</p></div><div class="footer-links"><div class="footer-column"><h4>Product</h4><a href="/#features">Features</a><a href="/pricing">Pricing</a><a href="https://docs.agent-relay.com/">Documentation</a><a href="/changelog">Changelog</a></div><div class="footer-column"><h4>Company</h4><a href="/about">About</a><a href="/blog">Blog</a><a href="/careers">Careers</a><a href="/contact">Contact</a></div><div class="footer-column"><h4>Legal</h4><a href="/privacy">Privacy</a><a href="/terms">Terms</a><a href="/security">Security</a></div></div></div><div class="footer-bottom"><p>© 2026 Agent Relay. All rights reserved.</p><div class="social-links"><a href="https://github.com/AgentWorkforce/relay" aria-label="GitHub"><svg width="20" height="20" viewBox="0 0 24 24" fill="currentColor"><path d="M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0024 12c0-6.63-5.37-12-12-12z"></path></svg></a><a href="https://twitter.com/agent_relay" aria-label="Twitter"><svg width="20" height="20" viewBox="0 0 24 24" fill="currentColor"><path d="M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z"></path></svg></a></div></div></footer></div><script src="/_next/static/chunks/webpack-dd93b81e2659669c.js" async=""></script><script>(self.__next_f=self.__next_f||[]).push([0]);self.__next_f.push([2,null])</script><script>self.__next_f.push([1,"1:HL[\"/_next/static/css/ad96af0f7a47b705.css\",\"style\"]\n2:HL[\"/_next/static/css/629d21db423c78a2.css\",\"style\"]\n"])</script><script>self.__next_f.push([1,"3:I[1815,[],\"\"]\n5:I[2147,[\"294\",\"static/chunks/app/blog/let-them-cook-multi-agent-orchestration/page-a58308f43557b908.js\"],\"\"]\n6:I[5092,[],\"\"]\n7:I[2023,[],\"\"]\n9:I[1584,[],\"\"]\na:[]\n"])</script><script>self.__next_f.push([1,"0:[\"$\",\"$L3\",null,{\"buildId\":\"Zy1iEBTs-RzAM6cw9InUm\",\"assetPrefix\":\"\",\"urlParts\":[\"\",\"blog\",\"let-them-cook-multi-agent-orchestration\"],\"initialTree\":[\"\",{\"children\":[\"blog\",{\"children\":[\"let-them-cook-multi-agent-orchestration\",{\"children\":[\"__PAGE__\",{}]}]}]},\"$undefined\",\"$undefined\",true],\"initialSeedData\":[\"\",{\"children\":[\"blog\",{\"children\":[\"let-them-cook-multi-agent-orchestration\",{\"children\":[\"__PAGE__\",{},[[\"$L4\",[\"$\",\"div\",null,{\"className\":\"landing-page static-page\",\"children\":[[\"$\",\"div\",null,{\"className\":\"landing-bg\",\"children\":[[\"$\",\"div\",null,{\"className\":\"grid-bg\",\"children\":[[\"$\",\"div\",null,{\"className\":\"grid-lines\"}],[\"$\",\"div\",null,{\"className\":\"grid-glow\"}]]}],[\"$\",\"div\",null,{\"className\":\"glow-orbs\",\"children\":[[\"$\",\"div\",null,{\"className\":\"orb orb-1\"}],[\"$\",\"div\",null,{\"className\":\"orb orb-2\"}],[\"$\",\"div\",null,{\"className\":\"orb orb-3\"}]]}]]}],[\"$\",\"nav\",null,{\"className\":\"nav scrolled\",\"children\":[\"$\",\"div\",null,{\"className\":\"nav-inner\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/\",\"className\":\"nav-logo\",\"children\":[[\"$\",\"svg\",null,{\"width\":28,\"height\":28,\"viewBox\":\"0 0 100 100\",\"fill\":\"none\",\"xmlns\":\"http://www.w3.org/2000/svg\",\"className\":\"\\n        transition-all duration-300\\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\\n        \\n      \",\"aria-label\":\"Agent Relay Logo\",\"role\":\"img\",\"children\":[[\"$\",\"path\",null,{\"d\":\"M30 80 L 50 20 L 70 80\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinejoin\":\"round\",\"strokeLinecap\":\"round\",\"fill\":\"none\"}],[\"$\",\"line\",null,{\"x1\":\"40\",\"y1\":\"50\",\"x2\":\"60\",\"y2\":\"50\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinecap\":\"round\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 L 50 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"opacity\":\"0.7\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 C 80 20 80 50 50 50 L 80 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"strokeLinejoin\":\"round\",\"fill\":\"none\",\"opacity\":\"0.7\"}]]}],[\"$\",\"span\",null,{\"className\":\"logo-text\",\"children\":\"Agent Relay\"}]]}],[\"$\",\"div\",null,{\"className\":\"nav-links\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/#demo\",\"children\":\"Demo\"}],[\"$\",\"a\",null,{\"href\":\"/#features\",\"children\":\"Features\"}],[\"$\",\"a\",null,{\"href\":\"/pricing\",\"children\":\"Pricing\"}],[\"$\",\"a\",null,{\"href\":\"https://docs.agent-relay.com/\",\"className\":\"nav-docs\",\"children\":\"Docs\"}]]}],[\"$\",\"div\",null,{\"className\":\"nav-actions\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/login\",\"className\":\"btn-ghost\",\"children\":\"Sign In\"}],[\"$\",\"a\",null,{\"href\":\"/signup\",\"className\":\"btn-primary\",\"children\":\"Get Started\"}]]}]]}]}],[\"$\",\"main\",null,{\"children\":[[\"$\",\"section\",null,{\"className\":\"static-hero\",\"children\":[\"$\",\"div\",null,{\"className\":\"static-hero-content\",\"children\":[[\"$\",\"h1\",null,{\"children\":[\"$\",\"a\",null,{\"href\":\"/blog\",\"style\":{\"color\":\"inherit\",\"textDecoration\":\"none\"},\"children\":\"Blog\"}]}],[\"$\",\"p\",null,{\"className\":\"static-subtitle\",\"children\":\"News, tutorials, and insights from the Agent Relay team.\"}],\"$undefined\"]}]}],[\"$\",\"section\",null,{\"className\":\"static-content\",\"children\":[\"$\",\"div\",null,{\"className\":\"content-container\",\"children\":[[\"$\",\"article\",null,{\"className\":\"blog-post\",\"children\":[[\"$\",\"img\",null,{\"src\":\"/blog/let-them-cook-multi-agent-orchestration.svg\",\"alt\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\",\"style\":{\"width\":\"100%\",\"height\":\"auto\",\"borderRadius\":\"12px\",\"marginBottom\":\"32px\"}}],[\"$\",\"div\",null,{\"className\":\"blog-post-header\",\"children\":[[\"$\",\"div\",null,{\"className\":\"blog-meta\",\"children\":[[\"$\",\"span\",null,{\"className\":\"blog-category\",\"children\":\"Engineering\"}],[\"$\",\"span\",null,{\"children\":\"February 4, 2026\"}],[\"$\",\"span\",null,{\"className\":\"blog-author\",\"children\":[\"by \",[\"$\",\"a\",null,{\"href\":\"https://x.com/khaliqgant\",\"target\":\"_blank\",\"rel\":\"noopener noreferrer\",\"children\":\"Khaliq Gant\"}]]}]]}],[\"$\",\"h2\",null,{\"className\":\"blog-post-title\",\"children\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\"}],[\"$\",\"p\",null,{\"className\":\"blog-post-subtitle\",\"children\":\"What I learned watching AI agents coordinate, communicate, and occasionally fall apart\"}]]}],[\"$\",\"div\",null,{\"className\":\"blog-post-content\",\"children\":[[\"$\",\"p\",\"0\",{\"dangerouslySetInnerHTML\":{\"__html\":\"I've been building \u003ca href=\\\"https://github.com/AgentWorkforce/relay\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eAgent Relay\u003c/a\u003e using Agent Relay. Agents coordinating to build the tool that lets them coordinate. It sounds recursive because it is. I took some time to jot down some thoughts about multi-agent orchestration from the past few weeks where I've spoken to agents more than I've spoken to my wife 😳.\"}}],[\"$\",\"p\",\"1\",{\"dangerouslySetInnerHTML\":{\"__html\":\"For the past six weeks, I've been deep in this world. Agent Relay is an open-source communication layer allowing any CLI tool (Claude, Cursor, OpenCode, Gemini) to communicate efficiently and seamlessly. And it's rumored that Claude Code is coming out with first-party support for agent swarms:\"}}],[\"$\",\"div\",\"2\",{\"className\":\"tweet-embed\",\"style\":{\"margin\":\"24px 0\",\"display\":\"flex\",\"justifyContent\":\"center\"},\"children\":[\"$\",\"blockquote\",null,{\"className\":\"twitter-tweet\",\"data-theme\":\"dark\",\"children\":[\"$\",\"a\",null,{\"href\":\"https://twitter.com/nummanali/status/2014698883507462387\",\"children\":\"View tweet\"}]}]}],[\"$\",\"p\",\"3\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Am I worried about this? If I'm being completely honest, yes, a little bit. But my main feeling is that it's great to push multi-agent orchestration into the forefront of developer minds so the true power can be experienced and more best practices form around it.\"}}],[\"$\",\"h2\",\"4\",{\"children\":\"Key Takeaways\"}],[\"$\",\"ul\",\"5\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003e2-5 worker agents per Lead\u003c/strong\u003e is the sweet spot\"}}]}],[\"$\",\"ul\",\"6\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eClaude for coordination, Codex for deep work\u003c/strong\u003e (match the CLI to the role)\"}}]}],[\"$\",\"ul\",\"7\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003ePlanning is everything\u003c/strong\u003e (agents cut corners on vague specs)\"}}]}],[\"$\",\"ul\",\"8\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eShadow agents and reviewers\u003c/strong\u003e catch lazy work\"}}]}],[\"$\",\"ul\",\"9\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eStore trajectories\u003c/strong\u003e so future agents have context\"}}]}],[\"$\",\"p\",\"10\",{\"dangerouslySetInnerHTML\":{\"__html\":\"---\"}}],[\"$\",\"h2\",\"11\",{\"children\":\"Part 1: The Promise\"}],[\"$\",\"p\",\"12\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Multi-agent orchestration is a step change in how tasks get done. It puts agents front and center while the human takes a step back and just lets them cook. That doesn't mean the human will be completely removed. There are still bumps in the road to truly autonomous agent work, and the planning phase becomes one of the most crucial steps.\"}}],[\"$\",\"h3\",\"13\",{\"children\":\"What Multi-Agent Unlocks\"}],[\"$\",\"p\",\"14\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Having agents who can communicate with each other and coordinate on tasks is a huge unlock. Assigning \u003ca href=\\\"https://github.com/AgentWorkforce/relay/tree/main/.claude/agents\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eagent profiles\u003c/a\u003e similar to how human teams would organize has been a paradigm I have found success with. For instance:\"}}],[\"$\",\"ul\",\"15\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eLead\u003c/strong\u003e – coordinates the team and breaks down tasks\"}}]}],[\"$\",\"ul\",\"16\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eBackend\u003c/strong\u003e – implements server-side logic\"}}]}],[\"$\",\"ul\",\"17\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eBackendReviewer\u003c/strong\u003e – reviews backend code for quality\"}}]}],[\"$\",\"ul\",\"18\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eFrontendReviewer\u003c/strong\u003e – reviews frontend code for quality\"}}]}],[\"$\",\"ul\",\"19\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eTypeChecker\u003c/strong\u003e – ensures type safety across the codebase\"}}]}],[\"$\",\"ul\",\"20\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eTestWriter\u003c/strong\u003e – writes and maintains tests\"}}]}],[\"$\",\"ul\",\"21\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eDocumentationExpert\u003c/strong\u003e – handles docs and comments\"}}]}],[\"$\",\"p\",\"22\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Each agent assumes its role, can read the logs of other agents, and can message others to check their work, sanity-check their decisions, and hand off tasks in a coordinated manner.\"}}],[\"$\",\"h3\",\"23\",{\"children\":\"The Speed Improvement\"}],[\"$\",\"p\",\"24\",{\"dangerouslySetInnerHTML\":{\"__html\":\"The method that has worked well for me is creating a detailed spec upfront and then spawning a Lead agent. I give it the spec and tell it to assemble a team as it sees fit. The Lead then spawns agents accordingly. Because Agent Relay is CLI-agnostic, I make sure to mix Codex, Claude, OpenCode, Gemini, and Droid agents, assigning different models based on the role. A fast model like Haiku or Conductor for Lead roles, and for deeper technical tasks, Opus or GPT-5 Codex high.\"}}],[\"$\",\"p\",\"25\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Using this workflow, I've seen that not only does code quality increase, but the speed at which agents can pump out complex features is at least a \u003cstrong\u003e4-5X improvement\u003c/strong\u003e.\"}}],[\"$\",\"p\",\"26\",{\"dangerouslySetInnerHTML\":{\"__html\":\"I have been using Agent Relay to build itself on the cloud environment at \u003ca href=\\\"https://agent-relay.com\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eagent-relay.com\u003c/a\u003e and the pace of delivery has been mind-blowing.\"}}],[\"$\",\"p\",\"27\",{\"dangerouslySetInnerHTML\":{\"__html\":\"It hasn't been all good though...\"}}],[\"$\",\"p\",\"28\",{\"dangerouslySetInnerHTML\":{\"__html\":\"---\"}}],[\"$\",\"h2\",\"29\",{\"children\":\"Part 2: The Problems\"}],[\"$\",\"h3\",\"30\",{\"children\":\"Agents Are Sometimes Lazy\"}],[\"$\",\"p\",\"31\",{\"dangerouslySetInnerHTML\":{\"__html\":\"I've had instances where an agent swarm takes on a complicated feature and then the Lead excitedly declares everything done. On one build, the Lead proudly reported \\\"All 12 endpoints implemented!\\\" When I tested it, only 8 actually returned data. The rest were stubbed out with TODOs. This happened occasionally with single-agent sessions, but imagine it compounding across 5, 6, or 10 agents...\"}}],[\"$\",\"h3\",\"32\",{\"children\":\"Agents Get Overwhelmed and Die\"}],[\"$\",\"p\",\"33\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Having a Lead agent creates a single point of failure. If the swarm is large and chatty, the Lead receives a flood of messages from other agents plus queries from the human asking about status or redirecting work. This can overwhelm the Lead, causing it to enter an endless loop and eventually die or become completely unresponsive.\"}}],[\"$\",\"p\",\"34\",{\"dangerouslySetInnerHTML\":{\"__html\":\"---\"}}],[\"$\",\"h2\",\"35\",{\"children\":\"Part 3: The Playbook\"}],[\"$\",\"p\",\"36\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Here's what I've learned about making multi-agent orchestration actually work.\"}}],[\"$\",\"h3\",\"37\",{\"children\":\"Team Structure\"}],[\"$\",\"p\",\"38\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Team structure is critical. There's a magic ratio of Lead-to-worker agents that I haven't exactly figured out yet (it varies depending on roles) but I've had success with \u003cstrong\u003e2-5 worker agents per Lead\u003c/strong\u003e. A single Lead managing 10 agents usually becomes problematic.\"}}],[\"$\",\"h3\",\"39\",{\"children\":\"Communication Patterns by CLI\"}],[\"$\",\"p\",\"40\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Not all CLI agents communicate the same way, and taking this into account is beneficial when working with swarms.\"}}],[\"$\",\"p\",\"41\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eCodex\u003c/strong\u003e is great at heads-down work but doesn't communicate well. Once it's working, it's hard to interrupt. I've had Leads waiting 15+ minutes for a response, assuming the agent died, when Codex was just deep in implementation.\"}}],[\"$\",\"p\",\"42\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eClaude\u003c/strong\u003e communicates excellently and can be interrupted mid-task without issue. It naturally provides status updates and asks clarifying questions.\"}}],[\"$\",\"p\",\"43\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eCursor\u003c/strong\u003e (particularly Composer) is very fast and communicates well, making it good for rapid iteration.\"}}],[\"$\",\"p\",\"44\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eMy staffing rule of thumb:\u003c/strong\u003e Put Claude agents in Lead and Reviewer roles where communication is key. Put Codex agents on isolated implementation tasks where heads-down focus is more valuable than status updates. Use Cursor when you need speed and tight feedback loops.\"}}],[\"$\",\"p\",\"45\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Agent Relay allows users to define a \u003ccode\u003eteams.json\u003c/code\u003e (\u003ca href=\\\"https://docs.agent-relay.com/reference/configuration#project-configuration-files\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003edocs\u003c/a\u003e) that auto-spawns agents on start, so these staffing decisions can be codified and stay consistent across sessions:\"}}],[\"$\",\"pre\",\"46\",{\"className\":\"code-block\",\"data-lang\":\"json\",\"children\":[\"$\",\"code\",null,{\"children\":\"{\\n  \\\"team\\\": \\\"my-team\\\",\\n  \\\"autoSpawn\\\": true,\\n  \\\"agents\\\": [\\n    {\\n      \\\"name\\\": \\\"Coordinator\\\",\\n      \\\"cli\\\": \\\"claude\\\",\\n      \\\"role\\\": \\\"coordinator\\\",\\n      \\\"task\\\": \\\"Coordinate the team...\\\"\\n    },\\n    {\\n      \\\"name\\\": \\\"Developer\\\",\\n      \\\"cli\\\": \\\"codex\\\",\\n      \\\"task\\\": \\\"Implement features...\\\"\\n    }\\n  ]\\n}\"}]}],[\"$\",\"h3\",\"47\",{\"children\":\"Catching Lazy Work\"}],[\"$\",\"p\",\"48\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Agent Relay has a notion of a \u003ca href=\\\"https://docs.agent-relay.com/features/shadows#shadow-agents\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e\u003cstrong\u003eshadow agent\u003c/strong\u003e\u003c/a\u003e that helps quite a bit with this problem. Reviewer agents also typically catch shoddy work.\"}}],[\"$\",\"figure\",\"49\",{\"className\":\"blog-image\",\"style\":{\"margin\":\"32px 0\"},\"children\":[[\"$\",\"img\",null,{\"src\":\"/blog/agent-review-workflow.png\",\"alt\":\"Agent Relay dashboard showing Developer reporting completion to Coordinator, who then routes the work to ReviewerA for review\",\"style\":{\"width\":\"100%\",\"borderRadius\":\"8px\"}}],[\"$\",\"figcaption\",null,{\"style\":{\"textAlign\":\"center\",\"color\":\"var(--text-secondary)\",\"marginTop\":\"12px\",\"fontSize\":\"14px\",\"fontStyle\":\"italic\"},\"children\":\"Agent Relay dashboard showing Developer reporting completion to Coordinator, who then routes the work to ReviewerA for review\"}]]}],[\"$\",\"p\",\"50\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Additionally, layering in one of the many AI code review tools has been effective at catching minor issues.\"}}],[\"$\",\"h3\",\"51\",{\"children\":\"Continuity and Hooks\"}],[\"$\",\"p\",\"52\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Agent Relay has a \u003ca href=\\\"https://docs.agent-relay.com/guides/session-continuity#session-continuity\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e\u003ccode\u003econtinuity\u003c/code\u003e\u003c/a\u003e concept, largely borrowed from the \u003ca href=\\\"https://github.com/parcadei/Continuous-Claude-v3\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eContinuous Claude\u003c/a\u003e package by \u003ca href=\\\"https://x.com/parcadei\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@parcadei\u003c/a\u003e. This enables ephemeral agents that save their context periodically, get released, then spawn again and continue seamlessly by reading their saved state.\"}}],[\"$\",\"p\",\"53\",{\"dangerouslySetInnerHTML\":{\"__html\":\"If you want more granular control or access to agent lifecycle events, Agent Relay has an extensive \u003ca href=\\\"https://docs.agent-relay.com/guides/hooks\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003ehooks system\u003c/a\u003e with 7 different events:\"}}],[\"$\",\"div\",\"54\",{\"style\":{\"overflowX\":\"auto\",\"margin\":\"24px 0\"},\"children\":[\"$\",\"table\",null,{\"style\":{\"width\":\"100%\",\"borderCollapse\":\"collapse\",\"fontSize\":\"15px\"},\"children\":[[\"$\",\"thead\",null,{\"children\":[\"$\",\"tr\",null,{\"children\":[[\"$\",\"th\",\"0\",{\"style\":{\"padding\":\"12px 16px\",\"textAlign\":\"left\",\"borderBottom\":\"2px solid var(--border-subtle)\",\"fontWeight\":600},\"dangerouslySetInnerHTML\":{\"__html\":\"Hook\"}}],[\"$\",\"th\",\"1\",{\"style\":{\"padding\":\"12px 16px\",\"textAlign\":\"left\",\"borderBottom\":\"2px solid var(--border-subtle)\",\"fontWeight\":600},\"dangerouslySetInnerHTML\":{\"__html\":\"Description\"}}]]}]}],[\"$\",\"tbody\",null,{\"children\":[[\"$\",\"tr\",\"0\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonSessionStart\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent starts\"}}]]}],[\"$\",\"tr\",\"1\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonSessionEnd\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent stops\"}}]]}],[\"$\",\"tr\",\"2\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonOutput\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent produces output\"}}]]}],[\"$\",\"tr\",\"3\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonMessageReceived\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Receives relay message\"}}]]}],[\"$\",\"tr\",\"4\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonMessageSent\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Sends relay message\"}}]]}],[\"$\",\"tr\",\"5\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonIdle\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent inactive\"}}]]}],[\"$\",\"tr\",\"6\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonError\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Error occurs\"}}]]}]]}]]}]}],[\"$\",\"h3\",\"55\",{\"children\":\"Trajectories: Preserving Context\"}],[\"$\",\"p\",\"56\",{\"dangerouslySetInnerHTML\":{\"__html\":\"One unexpected unlock has been having agents store \u003cstrong\u003etrajectories\u003c/strong\u003e, a train of thought stored in logical chapters for a completed task. Inspired by \u003ca href=\\\"https://x.com/gergelyorosz/status/2002160432841097239\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003ethis thread\u003c/a\u003e from \u003ca href=\\\"https://x.com/GergelyOrosz\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@GergelyOrosz\u003c/a\u003e.\"}}],[\"$\",\"pre\",\"57\",{\"className\":\"code-block\",\"data-lang\":\"json\",\"children\":[\"$\",\"code\",null,{\"children\":\"{\\n  \\\"id\\\": \\\"traj_itn5hyej5mi6\\\",\\n  \\\"task\\\": { \\\"title\\\": \\\"Fix module resolution issues\\\" },\\n  \\\"status\\\": \\\"completed\\\",\\n  \\\"chapters\\\": [\\n    {\\n      \\\"title\\\": \\\"Work\\\",\\n      \\\"events\\\": [\\n        {\\n          \\\"type\\\": \\\"decision\\\",\\n          \\\"content\\\": \\\"Thread shadowMode through protocol layers\\\",\\n          \\\"raw\\\": { \\\"reasoning\\\": \\\"Devin review found fields were silently dropped...\\\" },\\n          \\\"significance\\\": \\\"high\\\"\\n        }\\n      ]\\n    }\\n  ],\\n  \\\"retrospective\\\": {\\n    \\\"summary\\\": \\\"Fixed issue by threading shadow options through all layers\\\",\\n    \\\"confidence\\\": 0.9\\n  }\\n}\"}]}],[\"$\",\"p\",\"58\",{\"dangerouslySetInnerHTML\":{\"__html\":\"The \u003ca href=\\\"https://github.com/AgentWorkforce/trajectories/\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eAgentWorkforce/trajectories\u003c/a\u003e repo provides a CLI tool that agents can easily understand. It becomes invaluable when an agent needs to revisit a previously-worked feature or investigate a bug. By finding the relevant trajectory, the agent gains instant context. These trajectories are also useful to humans reviewing the codebase.\"}}],[\"$\",\"h3\",\"59\",{\"children\":\"The Human's New Role\"}],[\"$\",\"p\",\"60\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Coordinating with multiple agents and seeing output fly in at rapid speed is exhilarating. Being able to remove myself as the bottleneck is a huge benefit. This necessitates that the \u003cstrong\u003eplanning phase is carefully and meticulously done\u003c/strong\u003e to ensure agents have well-defined tasks with edge cases thought out.\"}}],[\"$\",\"p\",\"61\",{\"dangerouslySetInnerHTML\":{\"__html\":\"It also means the \u003cstrong\u003ereview phase is paramount\u003c/strong\u003e. Having agents self-review and cross-review is an effective strategy.\"}}],[\"$\",\"p\",\"62\",{\"dangerouslySetInnerHTML\":{\"__html\":\"---\"}}],[\"$\",\"h2\",\"63\",{\"children\":\"Try It Yourself\"}],[\"$\",\"p\",\"64\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Want to experiment with multi-agent orchestration? You can \u003ca href=\\\"https://agent-relay.com\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eget started with Agent Relay Cloud\u003c/a\u003e or set it up locally in seconds.\"}}],[\"$\",\"p\",\"65\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Just tell your CLI agent to run:\"}}],[\"$\",\"pre\",\"66\",{\"className\":\"code-block\",\"data-lang\":\"bash\",\"children\":[\"$\",\"code\",null,{\"children\":\"curl -s https://raw.githubusercontent.com/AgentWorkforce/relay/main/docs/guide/agent-setup.md\"}]}],[\"$\",\"p\",\"67\",{\"dangerouslySetInnerHTML\":{\"__html\":\"The agent will read the setup guide and configure everything, then let it cook.\"}}],[\"$\",\"p\",\"68\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Check out the \u003ca href=\\\"https://docs.agent-relay.com\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003edocs\u003c/a\u003e for more details, or hit me on X with any thoughts or questions. I'm all for a discussion: \u003ca href=\\\"https://x.com/khaliqgant\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@khaliqgant\u003c/a\u003e\"}}],[\"$\",\"h2\",\"69\",{\"children\":\"References\"}],[\"$\",\"ul\",\"70\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ca href=\\\"https://cursor.com/blog/scaling-agents\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eCursor: Scaling Agents\u003c/a\u003e – An interesting perspective on how Cursor is thinking about agent scaling and the challenges involved\"}}]}],[\"$\",\"ul\",\"71\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ca href=\\\"https://x.com/pbteja1998/status/2017662163540971756\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@pbteja1998 on multi-agent swarms with OpenClaw\u003c/a\u003e – Exciting developments in how people are experimenting with multi-agent orchestration\"}}]}]]}]]}],[\"$\",\"hr\",null,{\"style\":{\"border\":\"none\",\"borderTop\":\"1px solid var(--border-subtle)\",\"margin\":\"48px 0\"}}],[\"$\",\"h2\",null,{\"children\":\"Stay Updated\"}],[\"$\",\"p\",null,{\"children\":[\"Follow us on \",[\"$\",\"a\",null,{\"href\":\"https://twitter.com/agent_relay\",\"target\":\"_blank\",\"rel\":\"noopener noreferrer\",\"children\":\"Twitter\"}],\" for the latest updates and more content.\"]}],[\"$\",\"$L5\",null,{\"src\":\"https://platform.twitter.com/widgets.js\",\"strategy\":\"lazyOnload\"}]]}]}]]}],[\"$\",\"footer\",null,{\"className\":\"footer\",\"children\":[[\"$\",\"div\",null,{\"className\":\"footer-inner\",\"children\":[[\"$\",\"div\",null,{\"className\":\"footer-brand\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/\",\"className\":\"footer-logo\",\"children\":[[\"$\",\"svg\",null,{\"width\":24,\"height\":24,\"viewBox\":\"0 0 100 100\",\"fill\":\"none\",\"xmlns\":\"http://www.w3.org/2000/svg\",\"className\":\"\\n        transition-all duration-300\\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\\n        \\n      \",\"aria-label\":\"Agent Relay Logo\",\"role\":\"img\",\"children\":[[\"$\",\"path\",null,{\"d\":\"M30 80 L 50 20 L 70 80\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinejoin\":\"round\",\"strokeLinecap\":\"round\",\"fill\":\"none\"}],[\"$\",\"line\",null,{\"x1\":\"40\",\"y1\":\"50\",\"x2\":\"60\",\"y2\":\"50\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinecap\":\"round\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 L 50 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"opacity\":\"0.7\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 C 80 20 80 50 50 50 L 80 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"strokeLinejoin\":\"round\",\"fill\":\"none\",\"opacity\":\"0.7\"}]]}],[\"$\",\"span\",null,{\"className\":\"logo-text\",\"children\":\"Agent Relay\"}]]}],[\"$\",\"p\",null,{\"children\":\"Orchestrate AI agents like a symphony.\"}]]}],[\"$\",\"div\",null,{\"className\":\"footer-links\",\"children\":[[\"$\",\"div\",null,{\"className\":\"footer-column\",\"children\":[[\"$\",\"h4\",null,{\"children\":\"Product\"}],[\"$\",\"a\",null,{\"href\":\"/#features\",\"children\":\"Features\"}],[\"$\",\"a\",null,{\"href\":\"/pricing\",\"children\":\"Pricing\"}],[\"$\",\"a\",null,{\"href\":\"https://docs.agent-relay.com/\",\"children\":\"Documentation\"}],[\"$\",\"a\",null,{\"href\":\"/changelog\",\"children\":\"Changelog\"}]]}],[\"$\",\"div\",null,{\"className\":\"footer-column\",\"children\":[[\"$\",\"h4\",null,{\"children\":\"Company\"}],[\"$\",\"a\",null,{\"href\":\"/about\",\"children\":\"About\"}],[\"$\",\"a\",null,{\"href\":\"/blog\",\"children\":\"Blog\"}],[\"$\",\"a\",null,{\"href\":\"/careers\",\"children\":\"Careers\"}],[\"$\",\"a\",null,{\"href\":\"/contact\",\"children\":\"Contact\"}]]}],[\"$\",\"div\",null,{\"className\":\"footer-column\",\"children\":[[\"$\",\"h4\",null,{\"children\":\"Legal\"}],[\"$\",\"a\",null,{\"href\":\"/privacy\",\"children\":\"Privacy\"}],[\"$\",\"a\",null,{\"href\":\"/terms\",\"children\":\"Terms\"}],[\"$\",\"a\",null,{\"href\":\"/security\",\"children\":\"Security\"}]]}]]}]]}],[\"$\",\"div\",null,{\"className\":\"footer-bottom\",\"children\":[[\"$\",\"p\",null,{\"children\":\"© 2026 Agent Relay. All rights reserved.\"}],[\"$\",\"div\",null,{\"className\":\"social-links\",\"children\":[[\"$\",\"a\",null,{\"href\":\"https://github.com/AgentWorkforce/relay\",\"aria-label\":\"GitHub\",\"children\":[\"$\",\"svg\",null,{\"width\":\"20\",\"height\":\"20\",\"viewBox\":\"0 0 24 24\",\"fill\":\"currentColor\",\"children\":[\"$\",\"path\",null,{\"d\":\"M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0024 12c0-6.63-5.37-12-12-12z\"}]}]}],[\"$\",\"a\",null,{\"href\":\"https://twitter.com/agent_relay\",\"aria-label\":\"Twitter\",\"children\":[\"$\",\"svg\",null,{\"width\":\"20\",\"height\":\"20\",\"viewBox\":\"0 0 24 24\",\"fill\":\"currentColor\",\"children\":[\"$\",\"path\",null,{\"d\":\"M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z\"}]}]}]]}]]}]]}]]}],[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/629d21db423c78a2.css\",\"precedence\":\"next\",\"crossOrigin\":\"$undefined\"}]]],null],null]},[null,[\"$\",\"$L6\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\",\"blog\",\"children\",\"let-them-cook-multi-agent-orchestration\",\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"errorScripts\":\"$undefined\",\"template\":[\"$\",\"$L7\",null,{}],\"templateStyles\":\"$undefined\",\"templateScripts\":\"$undefined\",\"notFound\":\"$undefined\",\"notFoundStyles\":\"$undefined\"}]],null]},[null,[\"$\",\"$L6\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\",\"blog\",\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"errorScripts\":\"$undefined\",\"template\":[\"$\",\"$L7\",null,{}],\"templateStyles\":\"$undefined\",\"templateScripts\":\"$undefined\",\"notFound\":\"$undefined\",\"notFoundStyles\":\"$undefined\"}]],null]},[[[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/ad96af0f7a47b705.css\",\"precedence\":\"next\",\"crossOrigin\":\"$undefined\"}]],[\"$\",\"html\",null,{\"lang\":\"en\",\"children\":[\"$\",\"body\",null,{\"children\":[\"$\",\"$L6\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"errorScripts\":\"$undefined\",\"template\":[\"$\",\"$L7\",null,{}],\"templateStyles\":\"$undefined\",\"templateScripts\":\"$undefined\",\"notFound\":[[\"$\",\"title\",null,{\"children\":\"404: This page could not be found.\"}],[\"$\",\"div\",null,{\"style\":{\"fontFamily\":\"system-ui,\\\"Segoe UI\\\",Roboto,Helvetica,Arial,sans-serif,\\\"Apple Color Emoji\\\",\\\"Segoe UI Emoji\\\"\",\"height\":\"100vh\",\"textAlign\":\"center\",\"display\":\"flex\",\"flexDirection\":\"column\",\"alignItems\":\"center\",\"justifyContent\":\"center\"},\"children\":[\"$\",\"div\",null,{\"children\":[[\"$\",\"style\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}\"}}],[\"$\",\"h1\",null,{\"className\":\"next-error-h1\",\"style\":{\"display\":\"inline-block\",\"margin\":\"0 20px 0 0\",\"padding\":\"0 23px 0 0\",\"fontSize\":24,\"fontWeight\":500,\"verticalAlign\":\"top\",\"lineHeight\":\"49px\"},\"children\":\"404\"}],[\"$\",\"div\",null,{\"style\":{\"display\":\"inline-block\"},\"children\":[\"$\",\"h2\",null,{\"style\":{\"fontSize\":14,\"fontWeight\":400,\"lineHeight\":\"49px\",\"margin\":0},\"children\":\"This page could not be found.\"}]}]]}]}]],\"notFoundStyles\":[]}]}]}]],null],null],\"couldBeIntercepted\":false,\"initialHead\":[null,\"$L8\"],\"globalErrorComponent\":\"$9\",\"missingSlots\":\"$Wa\"}]\n"])</script><script>self.__next_f.push([1,"8:[[\"$\",\"meta\",\"0\",{\"name\":\"viewport\",\"content\":\"width=device-width, initial-scale=1\"}],[\"$\",\"meta\",\"1\",{\"charSet\":\"utf-8\"}],[\"$\",\"title\",\"2\",{\"children\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration | Agent Relay Blog\"}],[\"$\",\"meta\",\"3\",{\"name\":\"description\",\"content\":\"Multi-agent orchestration is a step change in how tasks get done. After 6 weeks of building with agent swarms, here's what works, what breaks, and why the planning phase becomes everything.\"}],[\"$\",\"meta\",\"4\",{\"property\":\"og:title\",\"content\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\"}],[\"$\",\"meta\",\"5\",{\"property\":\"og:description\",\"content\":\"What I learned watching AI agents coordinate, communicate, and occasionally fall apart\"}],[\"$\",\"meta\",\"6\",{\"property\":\"og:url\",\"content\":\"https://agent-relay.com/blog/let-them-cook-multi-agent-orchestration\"}],[\"$\",\"meta\",\"7\",{\"property\":\"og:site_name\",\"content\":\"Agent Relay\"}],[\"$\",\"meta\",\"8\",{\"property\":\"og:image\",\"content\":\"https://agent-relay.com/blog/let-them-cook-multi-agent-orchestration.svg\"}],[\"$\",\"meta\",\"9\",{\"property\":\"og:image:width\",\"content\":\"1200\"}],[\"$\",\"meta\",\"10\",{\"property\":\"og:image:height\",\"content\":\"630\"}],[\"$\",\"meta\",\"11\",{\"property\":\"og:image:alt\",\"content\":\"Let Them Cook: Multi-Agent Orchestration - Agent Relay\"}],[\"$\",\"meta\",\"12\",{\"property\":\"og:type\",\"content\":\"article\"}],[\"$\",\"meta\",\"13\",{\"property\":\"article:published_time\",\"content\":\"2026-02-04\"}],[\"$\",\"meta\",\"14\",{\"property\":\"article:author\",\"content\":\"Khaliq Gant\"}],[\"$\",\"meta\",\"15\",{\"name\":\"twitter:card\",\"content\":\"summary_large_image\"}],[\"$\",\"meta\",\"16\",{\"name\":\"twitter:title\",\"content\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\"}],[\"$\",\"meta\",\"17\",{\"name\":\"twitter:description\",\"content\":\"What I learned watching AI agents coordinate, communicate, and occasionally fall apart\"}],[\"$\",\"meta\",\"18\",{\"name\":\"twitter:image\",\"content\":\"https://agent-relay.com/blog/let-them-cook-multi-agent-orchestration.svg\"}],[\"$\",\"link\",\"19\",{\"rel\":\"apple-touch-icon\",\"href\":\"/apple-icon.png?9e7a840704165ca6\",\"type\":\"image/png\",\"sizes\":\"256x256\"}]]\n"])</script><script>self.__next_f.push([1,"4:null\n"])</script></body></html>
+      " aria-label="Agent Relay Logo" role="img"><path d="M30 80 L 50 20 L 70 80" stroke="#00d9ff" stroke-width="5" stroke-linejoin="round" stroke-linecap="round" fill="none"></path><line x1="40" y1="50" x2="60" y2="50" stroke="#00d9ff" stroke-width="5" stroke-linecap="round"></line><path d="M50 20 L 50 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" opacity="0.7"></path><path d="M50 20 C 80 20 80 50 50 50 L 80 80" stroke="#00ffc8" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" fill="none" opacity="0.7"></path></svg><span class="logo-text">Agent Relay</span></a><p>Orchestrate AI agents like a symphony.</p></div><div class="footer-links"><div class="footer-column"><h4>Product</h4><a href="/#features">Features</a><a href="/pricing">Pricing</a><a href="https://docs.agent-relay.com/">Documentation</a><a href="/changelog">Changelog</a></div><div class="footer-column"><h4>Company</h4><a href="/about">About</a><a href="/blog">Blog</a><a href="/careers">Careers</a><a href="/contact">Contact</a></div><div class="footer-column"><h4>Legal</h4><a href="/privacy">Privacy</a><a href="/terms">Terms</a><a href="/security">Security</a></div></div></div><div class="footer-bottom"><p>© 2026 Agent Relay. All rights reserved.</p><div class="social-links"><a href="https://github.com/AgentWorkforce/relay" aria-label="GitHub"><svg width="20" height="20" viewBox="0 0 24 24" fill="currentColor"><path d="M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0024 12c0-6.63-5.37-12-12-12z"></path></svg></a><a href="https://twitter.com/agent_relay" aria-label="Twitter"><svg width="20" height="20" viewBox="0 0 24 24" fill="currentColor"><path d="M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z"></path></svg></a></div></div></footer></div><script src="/_next/static/chunks/webpack-dd93b81e2659669c.js" async=""></script><script>(self.__next_f=self.__next_f||[]).push([0]);self.__next_f.push([2,null])</script><script>self.__next_f.push([1,"1:HL[\"/_next/static/css/ad96af0f7a47b705.css\",\"style\"]\n2:HL[\"/_next/static/css/629d21db423c78a2.css\",\"style\"]\n"])</script><script>self.__next_f.push([1,"3:I[1815,[],\"\"]\n5:I[2147,[\"294\",\"static/chunks/app/blog/let-them-cook-multi-agent-orchestration/page-a58308f43557b908.js\"],\"\"]\n6:I[5092,[],\"\"]\n7:I[2023,[],\"\"]\n9:I[1584,[],\"\"]\na:[]\n"])</script><script>self.__next_f.push([1,"0:[\"$\",\"$L3\",null,{\"buildId\":\"MT7qdGNhzTqCgM6-WQcZL\",\"assetPrefix\":\"\",\"urlParts\":[\"\",\"blog\",\"let-them-cook-multi-agent-orchestration\"],\"initialTree\":[\"\",{\"children\":[\"blog\",{\"children\":[\"let-them-cook-multi-agent-orchestration\",{\"children\":[\"__PAGE__\",{}]}]}]},\"$undefined\",\"$undefined\",true],\"initialSeedData\":[\"\",{\"children\":[\"blog\",{\"children\":[\"let-them-cook-multi-agent-orchestration\",{\"children\":[\"__PAGE__\",{},[[\"$L4\",[\"$\",\"div\",null,{\"className\":\"landing-page static-page\",\"children\":[[\"$\",\"div\",null,{\"className\":\"landing-bg\",\"children\":[[\"$\",\"div\",null,{\"className\":\"grid-bg\",\"children\":[[\"$\",\"div\",null,{\"className\":\"grid-lines\"}],[\"$\",\"div\",null,{\"className\":\"grid-glow\"}]]}],[\"$\",\"div\",null,{\"className\":\"glow-orbs\",\"children\":[[\"$\",\"div\",null,{\"className\":\"orb orb-1\"}],[\"$\",\"div\",null,{\"className\":\"orb orb-2\"}],[\"$\",\"div\",null,{\"className\":\"orb orb-3\"}]]}]]}],[\"$\",\"nav\",null,{\"className\":\"nav scrolled\",\"children\":[\"$\",\"div\",null,{\"className\":\"nav-inner\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/\",\"className\":\"nav-logo\",\"children\":[[\"$\",\"svg\",null,{\"width\":28,\"height\":28,\"viewBox\":\"0 0 100 100\",\"fill\":\"none\",\"xmlns\":\"http://www.w3.org/2000/svg\",\"className\":\"\\n        transition-all duration-300\\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\\n        \\n      \",\"aria-label\":\"Agent Relay Logo\",\"role\":\"img\",\"children\":[[\"$\",\"path\",null,{\"d\":\"M30 80 L 50 20 L 70 80\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinejoin\":\"round\",\"strokeLinecap\":\"round\",\"fill\":\"none\"}],[\"$\",\"line\",null,{\"x1\":\"40\",\"y1\":\"50\",\"x2\":\"60\",\"y2\":\"50\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinecap\":\"round\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 L 50 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"opacity\":\"0.7\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 C 80 20 80 50 50 50 L 80 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"strokeLinejoin\":\"round\",\"fill\":\"none\",\"opacity\":\"0.7\"}]]}],[\"$\",\"span\",null,{\"className\":\"logo-text\",\"children\":\"Agent Relay\"}]]}],[\"$\",\"div\",null,{\"className\":\"nav-links\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/#demo\",\"children\":\"Demo\"}],[\"$\",\"a\",null,{\"href\":\"/#features\",\"children\":\"Features\"}],[\"$\",\"a\",null,{\"href\":\"/pricing\",\"children\":\"Pricing\"}],[\"$\",\"a\",null,{\"href\":\"https://docs.agent-relay.com/\",\"className\":\"nav-docs\",\"children\":\"Docs\"}]]}],[\"$\",\"div\",null,{\"className\":\"nav-actions\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/login\",\"className\":\"btn-ghost\",\"children\":\"Sign In\"}],[\"$\",\"a\",null,{\"href\":\"/signup\",\"className\":\"btn-primary\",\"children\":\"Get Started\"}]]}]]}]}],[\"$\",\"main\",null,{\"children\":[[\"$\",\"section\",null,{\"className\":\"static-hero\",\"children\":[\"$\",\"div\",null,{\"className\":\"static-hero-content\",\"children\":[[\"$\",\"h1\",null,{\"children\":[\"$\",\"a\",null,{\"href\":\"/blog\",\"style\":{\"color\":\"inherit\",\"textDecoration\":\"none\"},\"children\":\"Blog\"}]}],[\"$\",\"p\",null,{\"className\":\"static-subtitle\",\"children\":\"News, tutorials, and insights from the Agent Relay team.\"}],\"$undefined\"]}]}],[\"$\",\"section\",null,{\"className\":\"static-content\",\"children\":[\"$\",\"div\",null,{\"className\":\"content-container\",\"children\":[[\"$\",\"article\",null,{\"className\":\"blog-post\",\"children\":[[\"$\",\"img\",null,{\"src\":\"/blog/let-them-cook-multi-agent-orchestration.svg\",\"alt\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\",\"style\":{\"width\":\"100%\",\"height\":\"auto\",\"borderRadius\":\"12px\",\"marginBottom\":\"32px\"}}],[\"$\",\"div\",null,{\"className\":\"blog-post-header\",\"children\":[[\"$\",\"div\",null,{\"className\":\"blog-meta\",\"children\":[[\"$\",\"span\",null,{\"className\":\"blog-category\",\"children\":\"Engineering\"}],[\"$\",\"span\",null,{\"children\":\"February 4, 2026\"}],[\"$\",\"span\",null,{\"className\":\"blog-author\",\"children\":[\"by \",[\"$\",\"a\",null,{\"href\":\"https://x.com/khaliqgant\",\"target\":\"_blank\",\"rel\":\"noopener noreferrer\",\"children\":\"Khaliq Gant\"}]]}]]}],[\"$\",\"h2\",null,{\"className\":\"blog-post-title\",\"children\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\"}],[\"$\",\"p\",null,{\"className\":\"blog-post-subtitle\",\"children\":\"What I learned watching AI agents coordinate, communicate, and occasionally fall apart\"}]]}],[\"$\",\"div\",null,{\"className\":\"blog-post-content\",\"children\":[[\"$\",\"p\",\"0\",{\"dangerouslySetInnerHTML\":{\"__html\":\"I've been building \u003ca href=\\\"https://github.com/AgentWorkforce/relay\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eAgent Relay\u003c/a\u003e using Agent Relay. Agents coordinating to build the tool that lets them coordinate. It sounds recursive because it is. I took some time to jot down some thoughts about multi-agent orchestration from the past few weeks where I've spoken to agents more than I've spoken to my wife 😳.\"}}],[\"$\",\"p\",\"1\",{\"dangerouslySetInnerHTML\":{\"__html\":\"For the past six weeks, I've been deep in this world. Agent Relay is an open-source communication layer allowing any CLI tool (Claude, Cursor, OpenCode, Gemini) to communicate efficiently and seamlessly. And it's rumored that Claude Code is coming out with first-party support for agent swarms:\"}}],[\"$\",\"div\",\"2\",{\"className\":\"tweet-embed\",\"style\":{\"margin\":\"24px 0\",\"display\":\"flex\",\"justifyContent\":\"center\"},\"children\":[\"$\",\"blockquote\",null,{\"className\":\"twitter-tweet\",\"data-theme\":\"dark\",\"children\":[\"$\",\"a\",null,{\"href\":\"https://twitter.com/nummanali/status/2014698883507462387\",\"children\":\"View tweet\"}]}]}],[\"$\",\"p\",\"3\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Am I worried about this? If I'm being completely honest, yes, a little bit. But my main feeling is that it's great to push multi-agent orchestration into the forefront of developer minds so the true power can be experienced and more best practices form around it.\"}}],[\"$\",\"h2\",\"4\",{\"children\":\"Key Takeaways\"}],[\"$\",\"ul\",\"5\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003e2-5 worker agents per Lead\u003c/strong\u003e is the sweet spot\"}}]}],[\"$\",\"ul\",\"6\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eClaude for coordination, Codex for deep work\u003c/strong\u003e (match the CLI to the role)\"}}]}],[\"$\",\"ul\",\"7\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003ePlanning is everything\u003c/strong\u003e (agents cut corners on vague specs)\"}}]}],[\"$\",\"ul\",\"8\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eShadow agents and reviewers\u003c/strong\u003e catch lazy work\"}}]}],[\"$\",\"ul\",\"9\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eStore trajectories\u003c/strong\u003e so future agents have context\"}}]}],[\"$\",\"h2\",\"10\",{\"children\":\"Part 1: The Promise\"}],[\"$\",\"p\",\"11\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Multi-agent orchestration is a step change in how tasks get done. It puts agents front and center while the human takes a step back and just lets them cook. That doesn't mean the human will be completely removed. There are still bumps in the road to truly autonomous agent work, and the planning phase becomes one of the most crucial steps.\"}}],[\"$\",\"h3\",\"12\",{\"children\":\"What Multi-Agent Unlocks\"}],[\"$\",\"p\",\"13\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Having agents who can communicate with each other and coordinate on tasks is a huge unlock. Assigning \u003ca href=\\\"https://github.com/AgentWorkforce/relay/tree/main/.claude/agents\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eagent profiles\u003c/a\u003e similar to how human teams would organize has been a paradigm I have found success with. For instance:\"}}],[\"$\",\"ul\",\"14\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eLead\u003c/strong\u003e – coordinates the team and breaks down tasks\"}}]}],[\"$\",\"ul\",\"15\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eBackend\u003c/strong\u003e – implements server-side logic\"}}]}],[\"$\",\"ul\",\"16\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eBackendReviewer\u003c/strong\u003e – reviews backend code for quality\"}}]}],[\"$\",\"ul\",\"17\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eFrontendReviewer\u003c/strong\u003e – reviews frontend code for quality\"}}]}],[\"$\",\"ul\",\"18\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eTypeChecker\u003c/strong\u003e – ensures type safety across the codebase\"}}]}],[\"$\",\"ul\",\"19\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eTestWriter\u003c/strong\u003e – writes and maintains tests\"}}]}],[\"$\",\"ul\",\"20\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eDocumentationExpert\u003c/strong\u003e – handles docs and comments\"}}]}],[\"$\",\"p\",\"21\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Each agent assumes its role, can read the logs of other agents, and can message others to check their work, sanity-check their decisions, and hand off tasks in a coordinated manner.\"}}],[\"$\",\"h3\",\"22\",{\"children\":\"The Speed Improvement\"}],[\"$\",\"p\",\"23\",{\"dangerouslySetInnerHTML\":{\"__html\":\"The method that has worked well for me is creating a detailed spec upfront and then spawning a Lead agent. I give it the spec and tell it to assemble a team as it sees fit. The Lead then spawns agents accordingly. Because Agent Relay is CLI-agnostic, I make sure to mix Codex, Claude, OpenCode, Gemini, and Droid agents, assigning different models based on the role. A fast model like Haiku or Conductor for Lead roles, and for deeper technical tasks, Opus or GPT-5 Codex high.\"}}],[\"$\",\"p\",\"24\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Using this workflow, I've seen that not only does code quality increase, but the speed at which agents can pump out complex features is at least a \u003cstrong\u003e4-5X improvement\u003c/strong\u003e.\"}}],[\"$\",\"p\",\"25\",{\"dangerouslySetInnerHTML\":{\"__html\":\"I have been using Agent Relay to build itself on the cloud environment at \u003ca href=\\\"https://agent-relay.com\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eagent-relay.com\u003c/a\u003e and the pace of delivery has been mind-blowing.\"}}],[\"$\",\"p\",\"26\",{\"dangerouslySetInnerHTML\":{\"__html\":\"It hasn't been all good though...\"}}],[\"$\",\"h2\",\"27\",{\"children\":\"Part 2: The Problems\"}],[\"$\",\"h3\",\"28\",{\"children\":\"Agents Are Sometimes Lazy\"}],[\"$\",\"p\",\"29\",{\"dangerouslySetInnerHTML\":{\"__html\":\"I've had instances where an agent swarm takes on a complicated feature and then the Lead excitedly declares everything done. On one build, the Lead proudly reported \\\"All 12 endpoints implemented!\\\" When I tested it, only 8 actually returned data. The rest were stubbed out with TODOs. This happened occasionally with single-agent sessions, but imagine it compounding across 5, 6, or 10 agents...\"}}],[\"$\",\"h3\",\"30\",{\"children\":\"Agents Get Overwhelmed and Die\"}],[\"$\",\"p\",\"31\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Having a Lead agent creates a single point of failure. If the swarm is large and chatty, the Lead receives a flood of messages from other agents plus queries from the human asking about status or redirecting work. This can overwhelm the Lead, causing it to enter an endless loop and eventually die or become completely unresponsive.\"}}],[\"$\",\"h2\",\"32\",{\"children\":\"Part 3: The Playbook\"}],[\"$\",\"p\",\"33\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Here's what I've learned about making multi-agent orchestration actually work.\"}}],[\"$\",\"h3\",\"34\",{\"children\":\"Team Structure\"}],[\"$\",\"p\",\"35\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Team structure is critical. There's a magic ratio of Lead-to-worker agents that I haven't exactly figured out yet (it varies depending on roles) but I've had success with \u003cstrong\u003e2-5 worker agents per Lead\u003c/strong\u003e. A single Lead managing 10 agents usually becomes problematic.\"}}],[\"$\",\"h3\",\"36\",{\"children\":\"Communication Patterns by CLI\"}],[\"$\",\"p\",\"37\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Not all CLI agents communicate the same way, and taking this into account is beneficial when working with swarms.\"}}],[\"$\",\"p\",\"38\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eCodex\u003c/strong\u003e is great at heads-down work but doesn't communicate well. Once it's working, it's hard to interrupt. I've had Leads waiting 15+ minutes for a response, assuming the agent died, when Codex was just deep in implementation.\"}}],[\"$\",\"p\",\"39\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eClaude\u003c/strong\u003e communicates excellently and can be interrupted mid-task without issue. It naturally provides status updates and asks clarifying questions.\"}}],[\"$\",\"p\",\"40\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eCursor\u003c/strong\u003e (particularly Composer) is very fast and communicates well, making it good for rapid iteration.\"}}],[\"$\",\"p\",\"41\",{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003cstrong\u003eMy staffing rule of thumb:\u003c/strong\u003e Put Claude agents in Lead and Reviewer roles where communication is key. Put Codex agents on isolated implementation tasks where heads-down focus is more valuable than status updates. Use Cursor when you need speed and tight feedback loops.\"}}],[\"$\",\"p\",\"42\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Agent Relay allows users to define a \u003ccode\u003eteams.json\u003c/code\u003e (\u003ca href=\\\"https://docs.agent-relay.com/reference/configuration#project-configuration-files\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003edocs\u003c/a\u003e) that auto-spawns agents on start, so these staffing decisions can be codified and stay consistent across sessions:\"}}],[\"$\",\"pre\",\"43\",{\"className\":\"code-block\",\"data-lang\":\"json\",\"children\":[\"$\",\"code\",null,{\"children\":\"{\\n  \\\"team\\\": \\\"my-team\\\",\\n  \\\"autoSpawn\\\": true,\\n  \\\"agents\\\": [\\n    {\\n      \\\"name\\\": \\\"Coordinator\\\",\\n      \\\"cli\\\": \\\"claude\\\",\\n      \\\"role\\\": \\\"coordinator\\\",\\n      \\\"task\\\": \\\"Coordinate the team...\\\"\\n    },\\n    {\\n      \\\"name\\\": \\\"Developer\\\",\\n      \\\"cli\\\": \\\"codex\\\",\\n      \\\"task\\\": \\\"Implement features...\\\"\\n    }\\n  ]\\n}\"}]}],[\"$\",\"h3\",\"44\",{\"children\":\"Catching Lazy Work\"}],[\"$\",\"p\",\"45\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Agent Relay has a notion of a \u003ca href=\\\"https://docs.agent-relay.com/features/shadows#shadow-agents\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e\u003cstrong\u003eshadow agent\u003c/strong\u003e\u003c/a\u003e that helps quite a bit with this problem. Reviewer agents also typically catch shoddy work.\"}}],[\"$\",\"figure\",\"46\",{\"className\":\"blog-image\",\"style\":{\"margin\":\"32px 0\"},\"children\":[[\"$\",\"img\",null,{\"src\":\"/blog/agent-review-workflow.png\",\"alt\":\"Agent Relay dashboard showing Developer reporting completion to Coordinator, who then routes the work to ReviewerA for review\",\"style\":{\"width\":\"100%\",\"borderRadius\":\"8px\"}}],[\"$\",\"figcaption\",null,{\"style\":{\"textAlign\":\"center\",\"color\":\"var(--text-secondary)\",\"marginTop\":\"12px\",\"fontSize\":\"14px\",\"fontStyle\":\"italic\"},\"children\":\"Agent Relay dashboard showing Developer reporting completion to Coordinator, who then routes the work to ReviewerA for review\"}]]}],[\"$\",\"p\",\"47\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Additionally, layering in one of the many AI code review tools has been effective at catching minor issues.\"}}],[\"$\",\"h3\",\"48\",{\"children\":\"Continuity and Hooks\"}],[\"$\",\"p\",\"49\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Agent Relay has a \u003ca href=\\\"https://docs.agent-relay.com/guides/session-continuity#session-continuity\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e\u003ccode\u003econtinuity\u003c/code\u003e\u003c/a\u003e concept, largely borrowed from the \u003ca href=\\\"https://github.com/parcadei/Continuous-Claude-v3\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eContinuous Claude\u003c/a\u003e package by \u003ca href=\\\"https://x.com/parcadei\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@parcadei\u003c/a\u003e. This enables ephemeral agents that save their context periodically, get released, then spawn again and continue seamlessly by reading their saved state.\"}}],[\"$\",\"p\",\"50\",{\"dangerouslySetInnerHTML\":{\"__html\":\"If you want more granular control or access to agent lifecycle events, Agent Relay has an extensive \u003ca href=\\\"https://docs.agent-relay.com/guides/hooks\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003ehooks system\u003c/a\u003e with 7 different events:\"}}],[\"$\",\"div\",\"51\",{\"style\":{\"overflowX\":\"auto\",\"margin\":\"24px 0\"},\"children\":[\"$\",\"table\",null,{\"style\":{\"width\":\"100%\",\"borderCollapse\":\"collapse\",\"fontSize\":\"15px\"},\"children\":[[\"$\",\"thead\",null,{\"children\":[\"$\",\"tr\",null,{\"children\":[[\"$\",\"th\",\"0\",{\"style\":{\"padding\":\"12px 16px\",\"textAlign\":\"left\",\"borderBottom\":\"2px solid var(--border-subtle)\",\"fontWeight\":600},\"dangerouslySetInnerHTML\":{\"__html\":\"Hook\"}}],[\"$\",\"th\",\"1\",{\"style\":{\"padding\":\"12px 16px\",\"textAlign\":\"left\",\"borderBottom\":\"2px solid var(--border-subtle)\",\"fontWeight\":600},\"dangerouslySetInnerHTML\":{\"__html\":\"Description\"}}]]}]}],[\"$\",\"tbody\",null,{\"children\":[[\"$\",\"tr\",\"0\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonSessionStart\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent starts\"}}]]}],[\"$\",\"tr\",\"1\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonSessionEnd\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent stops\"}}]]}],[\"$\",\"tr\",\"2\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonOutput\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent produces output\"}}]]}],[\"$\",\"tr\",\"3\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonMessageReceived\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Receives relay message\"}}]]}],[\"$\",\"tr\",\"4\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonMessageSent\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Sends relay message\"}}]]}],[\"$\",\"tr\",\"5\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonIdle\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Agent inactive\"}}]]}],[\"$\",\"tr\",\"6\",{\"children\":[[\"$\",\"td\",\"0\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ccode\u003eonError\u003c/code\u003e\"}}],[\"$\",\"td\",\"1\",{\"style\":{\"padding\":\"10px 16px\",\"borderBottom\":\"1px solid var(--border-subtle)\"},\"dangerouslySetInnerHTML\":{\"__html\":\"Error occurs\"}}]]}]]}]]}]}],[\"$\",\"h3\",\"52\",{\"children\":\"Trajectories: Preserving Context\"}],[\"$\",\"p\",\"53\",{\"dangerouslySetInnerHTML\":{\"__html\":\"One unexpected unlock has been having agents store \u003cstrong\u003etrajectories\u003c/strong\u003e, a train of thought stored in logical chapters for a completed task. Inspired by \u003ca href=\\\"https://x.com/gergelyorosz/status/2002160432841097239\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003ethis thread\u003c/a\u003e from \u003ca href=\\\"https://x.com/GergelyOrosz\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@GergelyOrosz\u003c/a\u003e.\"}}],[\"$\",\"pre\",\"54\",{\"className\":\"code-block\",\"data-lang\":\"json\",\"children\":[\"$\",\"code\",null,{\"children\":\"{\\n  \\\"id\\\": \\\"traj_itn5hyej5mi6\\\",\\n  \\\"task\\\": { \\\"title\\\": \\\"Fix module resolution issues\\\" },\\n  \\\"status\\\": \\\"completed\\\",\\n  \\\"chapters\\\": [\\n    {\\n      \\\"title\\\": \\\"Work\\\",\\n      \\\"events\\\": [\\n        {\\n          \\\"type\\\": \\\"decision\\\",\\n          \\\"content\\\": \\\"Thread shadowMode through protocol layers\\\",\\n          \\\"raw\\\": { \\\"reasoning\\\": \\\"Devin review found fields were silently dropped...\\\" },\\n          \\\"significance\\\": \\\"high\\\"\\n        }\\n      ]\\n    }\\n  ],\\n  \\\"retrospective\\\": {\\n    \\\"summary\\\": \\\"Fixed issue by threading shadow options through all layers\\\",\\n    \\\"confidence\\\": 0.9\\n  }\\n}\"}]}],[\"$\",\"p\",\"55\",{\"dangerouslySetInnerHTML\":{\"__html\":\"The \u003ca href=\\\"https://github.com/AgentWorkforce/trajectories/\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eAgentWorkforce/trajectories\u003c/a\u003e repo provides a CLI tool that agents can easily understand. It becomes invaluable when an agent needs to revisit a previously-worked feature or investigate a bug. By finding the relevant trajectory, the agent gains instant context. These trajectories are also useful to humans reviewing the codebase.\"}}],[\"$\",\"h3\",\"56\",{\"children\":\"The Human's New Role\"}],[\"$\",\"p\",\"57\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Coordinating with multiple agents and seeing output fly in at rapid speed is exhilarating. Being able to remove myself as the bottleneck is a huge benefit. This necessitates that the \u003cstrong\u003eplanning phase is carefully and meticulously done\u003c/strong\u003e to ensure agents have well-defined tasks with edge cases thought out.\"}}],[\"$\",\"p\",\"58\",{\"dangerouslySetInnerHTML\":{\"__html\":\"It also means the \u003cstrong\u003ereview phase is paramount\u003c/strong\u003e. Having agents self-review and cross-review is an effective strategy.\"}}],[\"$\",\"h2\",\"59\",{\"children\":\"Try It Yourself\"}],[\"$\",\"p\",\"60\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Want to experiment with multi-agent orchestration? You can \u003ca href=\\\"https://agent-relay.com\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eget started with Agent Relay Cloud\u003c/a\u003e or set it up locally in seconds.\"}}],[\"$\",\"p\",\"61\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Just tell your CLI agent to run:\"}}],[\"$\",\"pre\",\"62\",{\"className\":\"code-block\",\"data-lang\":\"bash\",\"children\":[\"$\",\"code\",null,{\"children\":\"curl -s https://raw.githubusercontent.com/AgentWorkforce/relay/main/docs/guide/agent-setup.md\"}]}],[\"$\",\"p\",\"63\",{\"dangerouslySetInnerHTML\":{\"__html\":\"The agent will read the setup guide and configure everything, then let it cook.\"}}],[\"$\",\"p\",\"64\",{\"dangerouslySetInnerHTML\":{\"__html\":\"Check out the \u003ca href=\\\"https://docs.agent-relay.com\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003edocs\u003c/a\u003e for more details, or hit me on X with any thoughts or questions. I'm all for a discussion: \u003ca href=\\\"https://x.com/khaliqgant\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@khaliqgant\u003c/a\u003e\"}}],[\"$\",\"h2\",\"65\",{\"children\":\"References\"}],[\"$\",\"ul\",\"66\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ca href=\\\"https://cursor.com/blog/scaling-agents\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003eCursor: Scaling Agents\u003c/a\u003e – An interesting perspective on how Cursor is thinking about agent scaling and the challenges involved\"}}]}],[\"$\",\"ul\",\"67\",{\"children\":[\"$\",\"li\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"\u003ca href=\\\"https://x.com/pbteja1998/status/2017662163540971756\\\" target=\\\"_blank\\\" rel=\\\"noopener noreferrer\\\"\u003e@pbteja1998 on multi-agent swarms with OpenClaw\u003c/a\u003e – Exciting developments in how people are experimenting with multi-agent orchestration\"}}]}]]}]]}],[\"$\",\"hr\",null,{\"style\":{\"border\":\"none\",\"borderTop\":\"1px solid var(--border-subtle)\",\"margin\":\"48px 0\"}}],[\"$\",\"h2\",null,{\"children\":\"Stay Updated\"}],[\"$\",\"p\",null,{\"children\":[\"Follow us on \",[\"$\",\"a\",null,{\"href\":\"https://twitter.com/agent_relay\",\"target\":\"_blank\",\"rel\":\"noopener noreferrer\",\"children\":\"Twitter\"}],\" for the latest updates and more content.\"]}],[\"$\",\"$L5\",null,{\"src\":\"https://platform.twitter.com/widgets.js\",\"strategy\":\"lazyOnload\"}]]}]}]]}],[\"$\",\"footer\",null,{\"className\":\"footer\",\"children\":[[\"$\",\"div\",null,{\"className\":\"footer-inner\",\"children\":[[\"$\",\"div\",null,{\"className\":\"footer-brand\",\"children\":[[\"$\",\"a\",null,{\"href\":\"/\",\"className\":\"footer-logo\",\"children\":[[\"$\",\"svg\",null,{\"width\":24,\"height\":24,\"viewBox\":\"0 0 100 100\",\"fill\":\"none\",\"xmlns\":\"http://www.w3.org/2000/svg\",\"className\":\"\\n        transition-all duration-300\\n        drop-shadow-[0_0_8px_rgba(0,217,255,0.3)]\\n        \\n      \",\"aria-label\":\"Agent Relay Logo\",\"role\":\"img\",\"children\":[[\"$\",\"path\",null,{\"d\":\"M30 80 L 50 20 L 70 80\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinejoin\":\"round\",\"strokeLinecap\":\"round\",\"fill\":\"none\"}],[\"$\",\"line\",null,{\"x1\":\"40\",\"y1\":\"50\",\"x2\":\"60\",\"y2\":\"50\",\"stroke\":\"#00d9ff\",\"strokeWidth\":\"5\",\"strokeLinecap\":\"round\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 L 50 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"opacity\":\"0.7\"}],[\"$\",\"path\",null,{\"d\":\"M50 20 C 80 20 80 50 50 50 L 80 80\",\"stroke\":\"#00ffc8\",\"strokeWidth\":\"2.5\",\"strokeLinecap\":\"round\",\"strokeLinejoin\":\"round\",\"fill\":\"none\",\"opacity\":\"0.7\"}]]}],[\"$\",\"span\",null,{\"className\":\"logo-text\",\"children\":\"Agent Relay\"}]]}],[\"$\",\"p\",null,{\"children\":\"Orchestrate AI agents like a symphony.\"}]]}],[\"$\",\"div\",null,{\"className\":\"footer-links\",\"children\":[[\"$\",\"div\",null,{\"className\":\"footer-column\",\"children\":[[\"$\",\"h4\",null,{\"children\":\"Product\"}],[\"$\",\"a\",null,{\"href\":\"/#features\",\"children\":\"Features\"}],[\"$\",\"a\",null,{\"href\":\"/pricing\",\"children\":\"Pricing\"}],[\"$\",\"a\",null,{\"href\":\"https://docs.agent-relay.com/\",\"children\":\"Documentation\"}],[\"$\",\"a\",null,{\"href\":\"/changelog\",\"children\":\"Changelog\"}]]}],[\"$\",\"div\",null,{\"className\":\"footer-column\",\"children\":[[\"$\",\"h4\",null,{\"children\":\"Company\"}],[\"$\",\"a\",null,{\"href\":\"/about\",\"children\":\"About\"}],[\"$\",\"a\",null,{\"href\":\"/blog\",\"children\":\"Blog\"}],[\"$\",\"a\",null,{\"href\":\"/careers\",\"children\":\"Careers\"}],[\"$\",\"a\",null,{\"href\":\"/contact\",\"children\":\"Contact\"}]]}],[\"$\",\"div\",null,{\"className\":\"footer-column\",\"children\":[[\"$\",\"h4\",null,{\"children\":\"Legal\"}],[\"$\",\"a\",null,{\"href\":\"/privacy\",\"children\":\"Privacy\"}],[\"$\",\"a\",null,{\"href\":\"/terms\",\"children\":\"Terms\"}],[\"$\",\"a\",null,{\"href\":\"/security\",\"children\":\"Security\"}]]}]]}]]}],[\"$\",\"div\",null,{\"className\":\"footer-bottom\",\"children\":[[\"$\",\"p\",null,{\"children\":\"© 2026 Agent Relay. All rights reserved.\"}],[\"$\",\"div\",null,{\"className\":\"social-links\",\"children\":[[\"$\",\"a\",null,{\"href\":\"https://github.com/AgentWorkforce/relay\",\"aria-label\":\"GitHub\",\"children\":[\"$\",\"svg\",null,{\"width\":\"20\",\"height\":\"20\",\"viewBox\":\"0 0 24 24\",\"fill\":\"currentColor\",\"children\":[\"$\",\"path\",null,{\"d\":\"M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0024 12c0-6.63-5.37-12-12-12z\"}]}]}],[\"$\",\"a\",null,{\"href\":\"https://twitter.com/agent_relay\",\"aria-label\":\"Twitter\",\"children\":[\"$\",\"svg\",null,{\"width\":\"20\",\"height\":\"20\",\"viewBox\":\"0 0 24 24\",\"fill\":\"currentColor\",\"children\":[\"$\",\"path\",null,{\"d\":\"M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z\"}]}]}]]}]]}]]}]]}],[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/629d21db423c78a2.css\",\"precedence\":\"next\",\"crossOrigin\":\"$undefined\"}]]],null],null]},[null,[\"$\",\"$L6\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\",\"blog\",\"children\",\"let-them-cook-multi-agent-orchestration\",\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"errorScripts\":\"$undefined\",\"template\":[\"$\",\"$L7\",null,{}],\"templateStyles\":\"$undefined\",\"templateScripts\":\"$undefined\",\"notFound\":\"$undefined\",\"notFoundStyles\":\"$undefined\"}]],null]},[null,[\"$\",\"$L6\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\",\"blog\",\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"errorScripts\":\"$undefined\",\"template\":[\"$\",\"$L7\",null,{}],\"templateStyles\":\"$undefined\",\"templateScripts\":\"$undefined\",\"notFound\":\"$undefined\",\"notFoundStyles\":\"$undefined\"}]],null]},[[[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/ad96af0f7a47b705.css\",\"precedence\":\"next\",\"crossOrigin\":\"$undefined\"}]],[\"$\",\"html\",null,{\"lang\":\"en\",\"children\":[\"$\",\"body\",null,{\"children\":[\"$\",\"$L6\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"errorScripts\":\"$undefined\",\"template\":[\"$\",\"$L7\",null,{}],\"templateStyles\":\"$undefined\",\"templateScripts\":\"$undefined\",\"notFound\":[[\"$\",\"title\",null,{\"children\":\"404: This page could not be found.\"}],[\"$\",\"div\",null,{\"style\":{\"fontFamily\":\"system-ui,\\\"Segoe UI\\\",Roboto,Helvetica,Arial,sans-serif,\\\"Apple Color Emoji\\\",\\\"Segoe UI Emoji\\\"\",\"height\":\"100vh\",\"textAlign\":\"center\",\"display\":\"flex\",\"flexDirection\":\"column\",\"alignItems\":\"center\",\"justifyContent\":\"center\"},\"children\":[\"$\",\"div\",null,{\"children\":[[\"$\",\"style\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}\"}}],[\"$\",\"h1\",null,{\"className\":\"next-error-h1\",\"style\":{\"display\":\"inline-block\",\"margin\":\"0 20px 0 0\",\"padding\":\"0 23px 0 0\",\"fontSize\":24,\"fontWeight\":500,\"verticalAlign\":\"top\",\"lineHeight\":\"49px\"},\"children\":\"404\"}],[\"$\",\"div\",null,{\"style\":{\"display\":\"inline-block\"},\"children\":[\"$\",\"h2\",null,{\"style\":{\"fontSize\":14,\"fontWeight\":400,\"lineHeight\":\"49px\",\"margin\":0},\"children\":\"This page could not be found.\"}]}]]}]}]],\"notFoundStyles\":[]}]}]}]],null],null],\"couldBeIntercepted\":false,\"initialHead\":[null,\"$L8\"],\"globalErrorComponent\":\"$9\",\"missingSlots\":\"$Wa\"}]\n"])</script><script>self.__next_f.push([1,"8:[[\"$\",\"meta\",\"0\",{\"name\":\"viewport\",\"content\":\"width=device-width, initial-scale=1\"}],[\"$\",\"meta\",\"1\",{\"charSet\":\"utf-8\"}],[\"$\",\"title\",\"2\",{\"children\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration | Agent Relay Blog\"}],[\"$\",\"meta\",\"3\",{\"name\":\"description\",\"content\":\"Multi-agent orchestration is a step change in how tasks get done. After 6 weeks of building with agent swarms, here's what works, what breaks, and why the planning phase becomes everything.\"}],[\"$\",\"meta\",\"4\",{\"property\":\"og:title\",\"content\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\"}],[\"$\",\"meta\",\"5\",{\"property\":\"og:description\",\"content\":\"What I learned watching AI agents coordinate, communicate, and occasionally fall apart\"}],[\"$\",\"meta\",\"6\",{\"property\":\"og:url\",\"content\":\"https://agent-relay.com/blog/let-them-cook-multi-agent-orchestration\"}],[\"$\",\"meta\",\"7\",{\"property\":\"og:site_name\",\"content\":\"Agent Relay\"}],[\"$\",\"meta\",\"8\",{\"property\":\"og:image\",\"content\":\"https://agent-relay.com/blog/let-them-cook-multi-agent-orchestration.svg\"}],[\"$\",\"meta\",\"9\",{\"property\":\"og:image:width\",\"content\":\"1200\"}],[\"$\",\"meta\",\"10\",{\"property\":\"og:image:height\",\"content\":\"630\"}],[\"$\",\"meta\",\"11\",{\"property\":\"og:image:alt\",\"content\":\"Let Them Cook: Multi-Agent Orchestration - Agent Relay\"}],[\"$\",\"meta\",\"12\",{\"property\":\"og:type\",\"content\":\"article\"}],[\"$\",\"meta\",\"13\",{\"property\":\"article:published_time\",\"content\":\"2026-02-04\"}],[\"$\",\"meta\",\"14\",{\"property\":\"article:author\",\"content\":\"Khaliq Gant\"}],[\"$\",\"meta\",\"15\",{\"name\":\"twitter:card\",\"content\":\"summary_large_image\"}],[\"$\",\"meta\",\"16\",{\"name\":\"twitter:title\",\"content\":\"Let Them Cook: Lessons from 6 Weeks of Multi-Agent Orchestration\"}],[\"$\",\"meta\",\"17\",{\"name\":\"twitter:description\",\"content\":\"What I learned watching AI agents coordinate, communicate, and occasionally fall apart\"}],[\"$\",\"meta\",\"18\",{\"name\":\"twitter:image\",\"content\":\"https://agent-relay.com/blog/let-them-cook-multi-agent-orchestration.svg\"}],[\"$\",\"link\",\"19\",{\"rel\":\"apple-touch-icon\",\"href\":\"/apple-icon.png?9e7a840704165ca6\",\"type\":\"image/png\",\"sizes\":\"256x256\"}]]\n"])</script><script>self.__next_f.push([1,"4:null\n"])</script></body></html>