[{"data":1,"prerenderedAt":327},["ShallowReactive",2],{"project-46990733":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":29,"sentiment":30,"stance_mix":31,"scored_comments":36,"aspects":37,"strip":44,"post_text":46,"what":47,"said":89,"maker_in_thread":194,"shape":205,"grid":208,"attempts":289,"same_week":290,"maker_shows":45,"maker_hidden":319,"reader_version":320,"card":321},"46990733","lean-collab","20+ Claude Code agents coordinating on real work (open source)","title","Show HN: 20+ Claude Code agents coordinating on real work (open source)","https:\u002F\u002Fgithub.com\u002Fmutable-state-inc\u002Flean-collab","github.com","austinbaggio",1770913417,2026,53,39,1170,4366,27,"notable","ai","multi-agent orchestration","cli",[24,25,26,27,28],"multi-agent","claude code","agent orchestration","lean proofs","llm agents","curious",-0.036,{"positive":32,"neutral":33,"mixed":34,"critical":35},4,10,0,5,19,{"idea":38,"execution":40,"design":42,"docs":43},{"positive":32,"critical":39},2,{"positive":34,"critical":41},1,{"positive":34,"critical":41},{"positive":34,"critical":34},[34,34,41,34,34,41,34,34,34,45,34,45,34,34,35,35,34,35,34,45,34,34,41,34,45,34,35,34,45,34,45,45,45,34,41,41,34,45,45],3,"Single-agent LLMs suck at long-running complex tasks.\n\nWe’ve open-sourced a multi-agent orchestrator that we’ve been using to handle long-running LLM tasks. We found that single LLM agents tend to stall, loop, or generate non-compiling code, so we built a harness for agents to coordinate over shared context while work is in progress.\n\nHow it works:\n1. Orchestrator agent that manages task decomposition\n2. Sub-agents for parallel work\n3. Subscriptions to task state and progress\n4. Real-time sharing of intermediate discoveries between agents\n\nWe tested this on a Putnam-level math problem, but the",{"problem":48,"audience":57,"pricing":57,"motivation":58,"time_to_build":57,"stack":65,"openness":87,"stage":88,"team_size":88},{"value":49,"source":50,"evidence":51},"Single LLM agents stall, loop, or generate non-compiling code on long-running complex tasks","maker",[52,54],{"id":53},"post",{"id":55,"by":11,"text":56},"46990987","All of the above. The most frustrating one with the Putnam example with Claude was generating solutions that obviously didn't compile. This feels like plan collapse- not…",null,{"value":59,"source":50,"evidence":60},"Single agents fail on long tasks; coordinating sub-agents over shared context scopes limited context windows",[61,62],{"id":53},{"id":63,"by":11,"text":64},"46996604","For most tasks, I agree. One agent with a good harness wins. The case for multiple agents is when the context required to solve the problem exceeds what one agent can hold. This…",[66,70,78],{"value":67,"source":50,"evidence":68},"Claude Code skill",[69],{"id":53},{"value":71,"source":72,"evidence":73},"Lean","commenter",[74],{"id":75,"by":76,"text":77},"46991389","raphaelmolly8","The Lean angle here is really interesting: most multi-agent demos dodge hard verification, but tying each agent’s output to makes the feedback loop objective. Curious how you’re…",{"value":79,"source":50,"evidence":80},"Ensue shared memory layer (key-value store with pub\u002Fsub and embeddings)",[81,84],{"id":82,"by":11,"text":83},"46995918","I'm using \"RAM\" loosely, meaning working memory here. In practice, it's a key-value store with pub\u002Fsub stored on our shared memory layer, Ensue. Agents write structured state to…",{"id":85,"by":11,"text":86},"46991678","Yeah we're using Ensue since it already handles the annoying infra pieces you’d otherwise have to build to make this work (shared task state + updates, event…","open_source","unknown",{"praise":90,"objections":102,"requests":125,"questions":150,"alternatives":169},[91,98],{"value":92,"source":72,"evidence":93},"Clean approach of bounded blast radius and code-encoded policy",[94],{"id":95,"by":96,"text":97},"46993348","visarga","Great work! I like the approach of maximum freedom inside bounded blast radius and how you use code to encode policy.",{"value":99,"source":72,"evidence":100},"Lean verification makes the feedback loop objective",[101],{"id":75,"by":76,"text":77},[103,114,118],{"value":104,"source":72,"evidence":105},"One well-tuned agent is enough for most tasks; multi-agent adds overhead",[106,110],{"id":107,"by":108,"text":109},"46994369","gck1","Even Anthropic research articles consistently demonstrate they themselves use one agent, and just tune the harness around it. I ignore all Skills, MCPs, and view all of these as…",{"id":111,"by":112,"text":113},"47000032","giancarlostoro","Thats the other thing, you hit the nail on the head, I dont want 20 agents unless they're doing research and scouring code. Claude can do that just fine. I want Claude Code doing…",{"value":115,"source":72,"evidence":116},"Hard to observe many agents at once",[117],{"id":107,"by":108,"text":109},{"value":119,"source":72,"evidence":120},"Requires a proprietary Ensue API key",[121],{"id":122,"by":123,"text":124},"46991576","slopusila","seems like it requires an API key to your proprietary Ensue memory system",[126,133,143],{"value":127,"source":72,"evidence":128},"Add clearer signup and API key recovery flow",[129],{"id":130,"by":131,"text":132},"46992238","yodon","When you get a chance to work on your login flow, I recommend giving users an opportunity to request the key rather than automatically showing it once only on the first screen. I…",{"value":134,"source":50,"evidence":135},"Suggest workloads to test next",[136,139],{"id":137,"by":11,"text":138},"46991461","Math proofs are really easy to run with this specific harness. Our next experiments are going to be bigger, think full code base refactors. We're working on applying RLM to…",{"id":140,"by":141,"text":142},"46991113","christinetyip","Cool, what’s a good first task to try this on where it’s likely to beat a single agent?",{"value":144,"source":72,"evidence":145},"Offline open-model setup for VPS",[146],{"id":147,"by":148,"text":149},"47000204","hsdev","Impressive! Are there any boilerplates that people know of for running something similar to this using open offline models? Would be cool to run this (or a single agent version)…",[151,158,162],{"value":152,"source":72,"evidence":153},"How does progress subscription work",[154],{"id":155,"by":156,"text":157},"46991168","zmanian","“How does progress subscription work — are agents watching specific signals (test failures, TODO list, build status), or just a global feed?”",{"value":159,"source":72,"evidence":160},"How are goal claim conflicts handled",[161],{"id":75,"by":76,"text":77},{"value":163,"source":72,"evidence":164},"Is shared memory RAM or a markdown file",[165],{"id":166,"by":167,"text":168},"46994728","arbol","So is this shared memory as in RAM or a markdown file that they update with their statuses?",[170,177,183,187],{"value":171,"source":72,"evidence":172},"Beads",[173,176],{"id":174,"by":112,"text":175},"46993598","I feel like there's two camps: * Throw more agents * Use something like Beads I'm in the latter, I don't have infinite resources, I'd rather stick to one agent and optimize what…",{"id":111,"by":112,"text":113},{"value":178,"source":72,"evidence":179},"GasTown",[180],{"id":181,"by":131,"text":182},"46991636","For the moment, researching multi-agent orchestration. At first glance, your work looks among the best in class of published work I've seen. Particularly interested to understand…",{"value":184,"source":72,"evidence":185},"Claude Code swarms",[186],{"id":181,"by":131,"text":182},{"value":188,"source":72,"evidence":189},"Ralph loop",[190],{"id":191,"by":192,"text":193},"46990954","miligauss","It's a more of a black box with claude, at least with this you see the proof strategy and mistakes made by the model when it decomposes the problem. I think instead of Ralph…",{"comments":195,"answered":196,"key":197},15,8,[198,201,202],{"id":199,"text":200},"46993660","Yeah I have seen those camps too. I think there will always be a set of problems that have complexity, measured by amount of context required to be kept in working ram, that need…",{"id":63,"text":64},{"id":203,"text":204},"46996524","I think about this with the analogue of MoE a lot. Essentially, a decision routing process, and similar to having expert submodels, you have a human in the loop or decision…",{"commenters":36,"max_depth":206,"maker_replies":195,"first_comment_mins":196,"lifespan_hours":207},6,101,[209,211,213,215,219,220,221,224,227,229,232,234,237,240,242,244,245,247,248,250,253,255,257,259,262,265,267,268,270,271,272,273,275,277,278,282,284,285,287],[174,34,34,210,112,34],"other\u002Fneutral\u002Fnone",[199,34,34,212,11,41],"project\u002Fneutral\u002Fidea",[107,34,41,214,108,34],"project\u002Fcritical\u002Fidea",[216,34,34,217,218,34],"46995559","commenter\u002Fneutral\u002Fnone","Kiboneu",[63,34,34,212,11,41],[111,34,41,214,112,34],[222,41,34,210,223,34],"46994638","raniazyane",[225,41,34,210,226,34],"46995079","joshribakoff",[203,41,34,228,11,41],"project\u002Fneutral\u002Fnone",[230,39,45,228,231,34],"46990828","clairekart",[55,39,34,233,11,41],"project\u002Fneutral\u002Fexecution",[235,39,45,233,236,34],"46991447","Atotalnoob",[238,45,34,210,239,34],"47011393","killbot_2000",[241,45,34,217,11,41],"47040783",[191,32,35,243,192,34],"project\u002Fpositive\u002Fidea",[75,32,35,243,76,34],[246,32,34,233,11,41],"46992632",[95,35,35,243,96,34],[249,35,34,212,11,41],"46993607",[251,206,45,252,131,34],"46991804","project\u002Fneutral\u002Fdesign",[254,206,34,252,11,41],"46991953",[256,206,34,252,11,41],"46991965",[130,206,41,258,131,34],"project\u002Fcritical\u002Fdesign",[260,206,34,261,11,41],"46992507","project\u002Fneutral\u002Fdocs",[263,264,45,261,131,34],"46990974",7,[266,264,34,261,11,41],"46991348",[181,264,35,243,131,34],[269,264,34,212,11,41],"46992454",[166,264,45,252,167,34],[82,264,34,252,11,41],[147,196,45,228,148,34],[140,274,45,228,141,34],9,[276,274,45,228,192,34],"46991457",[137,274,34,228,11,41],[279,274,41,280,281,34],"46997083","project\u002Fcritical\u002Fexecution","dimitri-vs",[122,33,41,283,123,34],"project\u002Fcritical\u002Fnone",[85,33,34,228,11,41],[155,286,45,233,156,34],11,[288,286,45,233,192,34],"46991431",[],[291,295,300,305,310,314],{"id":292,"name":293,"points":294,"category":20},"47031334","Jemini",488,{"id":296,"name":297,"points":298,"category":299},"47051852","AsteroidOS 2.0",470,"hardware",{"id":301,"name":302,"points":303,"category":304},"47026186","Microgpt",282,"learning_science",{"id":306,"name":307,"points":308,"category":309},"47040375","FreeFlow",277,"productivity",{"id":311,"name":312,"points":313,"category":304},"47008163","Data Engineering Book",251,{"id":315,"name":316,"points":317,"category":318},"47011567","SQL-tap",233,"dev_tools",false,"v2",{"id":322,"name":5,"line":6,"ai":319,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":323,"top":18,"kw":324,"maker":11,"domain":10,"tier":19,"attempts":34,"strip":325,"url":9,"mix":326,"praise":92,"objection":104,"sub":21},46990733,-0.04,[24,25,26,27,28],"001001000303005505030010305030333011033",[32,33,34,35],1794036693094]