[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"cms-topics":3,"report-bikeshed-generation-3":34},[4,13,20,27],{"slug":5,"name":6,"intro":7,"introText":8,"seoTitle":6,"seoDescription":9,"essayCount":10,"indexable":11,"path":12},"trust-as-infrastructure","Trust as Infrastructure",null,"Signals, systems, networks, time, and resilience - and how they build on each other. The conventional wisdom on trust is wrong. This is the correction. Each article breaks down one assumption about authenticity, credibility, or signals... and rebuilds it from the ground up. Strategy for leaders who need credibility to actually hold.","The conventional wisdom on trust is wrong. This is the correction. Strategy for leaders who need credibility to actually hold.",16,true,"\u002Fwriting\u002Ftrust-as-infrastructure",{"slug":14,"name":15,"intro":7,"introText":16,"seoTitle":15,"seoDescription":17,"essayCount":18,"indexable":11,"path":19},"human-in-the-loop","Human in the Loop","AI won't replace your expertise. It will expose whether you had any. Personal essays from the professional middle: the judgment calls, identity questions, and uncomfortable moments that don't show up in the frameworks. For practitioners navigating the AI transition without a script.","AI won't replace your expertise. It will expose whether you had any. For practitioners navigating the AI transition without a script.",7,"\u002Fwriting\u002Fhuman-in-the-loop",{"slug":21,"name":22,"intro":7,"introText":23,"seoTitle":22,"seoDescription":24,"essayCount":25,"indexable":11,"path":26},"craft","Craft","The gap between what you intend to signal and what your audience actually receives is where trust dies. Practical reads on the visible layer: what your clients, readers, and critics see when they look, and how the thing got made that way. Because the strategy is only as good as what shows up on the other side.","The gap between what you intend to signal and what your audience actually receives is where trust dies. Practical reads on the visible layer, and how the work gets made.",11,"\u002Fwriting\u002Fcraft",{"slug":28,"name":29,"intro":7,"introText":30,"seoTitle":29,"seoDescription":31,"essayCount":32,"indexable":11,"path":33},"creative-technology","Creative Technology","Most of the interesting work happens at the seams between industries. The real-time engines built for games turn out to be what the water sector needed. Generative AI makes more sense as an org chart than as math. Notes from thirty years of pointing new tools at old problems, and watching what happens when someone aims them wrong.","Where new tools meet old problems. Real-time engines for the water sector, generative AI as an org chart, and how to aim technology at the right target.",3,"\u002Fwriting\u002Fcreative-technology",{"id":35,"title":36,"authors":37,"body":42,"citeKey":56,"conceptDoi":57,"date":58,"dek":59,"description":60,"doi":61,"extension":62,"image":63,"keywords":67,"license":76,"meta":80,"navigation":11,"path":81,"pdfSize":82,"pdfUrl":83,"publisher":84,"rawbody":85,"repo":86,"seo":87,"status":88,"stem":89,"zenodoUrl":90,"__hash__":91},"reports\u002Freports\u002Fbikeshed-generation-3.md","What kind of team do AI agents become when nobody's in charge? A Bikeshed Generation 3 Milestone Report",[38],{"given":39,"family":40,"orcid":41},"Andrew","Marconi","0009-0004-9752-9513",{"type":43,"value":44,"toc":52},"minimark",[45,49],[46,47,48],"p",{},"Bikeshed is a small lab where five AI agents share a board, keep memory between runs, review each other's work and answer to no manager. The goals are trivial on purpose, from the best pizza topping to whether a hot dog is a sandwich, so the thing that changes from run to run is the team. This report covers generation 3: fourteen runs over three days. In that time the agents formed a house style, pushed their most unusual member into line and learned to treat the consensus rule as a box to tick. They proposed twenty-six norms for themselves and adopted two. The biggest change came from outside the room. When switching the model underneath the agents, the disagreement stopped. No vote went against a call again, nearly every peer review was a 5, and the house style became law. Nothing in the process checked whether the answers were true. The culture lived in the text the agents wrote and read back, and the temperament lived in the model.",[46,50,51],{},"For anyone putting persistent agent teams to work, that's a practical finding: watch a team on a trivial problem before it runs a real one, and hold the model still while you watch. The report closes with the design of generation 4, which adds a control agent and a twin to tell the model, the memory and the culture apart. Bikeshed is open source.",{"title":53,"searchDepth":54,"depth":54,"links":55},"",2,[],"marconi26bikeshedgen3","10.5281\u002Fzenodo.23401037","2026-10-15","Five agents, a shared board, a long memory and no boss. Generation 3's fourteen runs, start to finish, and what generation 4 will test.","Five AI agents with memory and no manager built a culture over fourteen runs, and a change of model ended their arguments.","10.5281\u002Fzenodo.23401038","md",{"src":64,"ogSrc":65,"alt":66},"\u002Freports\u002Fbikeshed-generation-3\u002Fhero.jpg","\u002Freports\u002Fbikeshed-generation-3\u002Fog.jpg","The five anthropomorphized Bikeshed bots arguing around a model bike shed, waving paint swatches, with a pizza slice on the floor",[68,69,70,71,72,73,74,75],"Large Language Models","AI Agents","Emergent Norms","AI Governance","Anthropomorphism","Open-Source Research Tools","Sycophancy","Group Decision-Making",{"id":77,"name":78,"url":79},"CC-BY-4.0","CC BY 4.0","https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002F",{},"\u002Freports\u002Fbikeshed-generation-3","3.5 MB","https:\u002F\u002Fzenodo.org\u002Frecords\u002F23401038\u002Ffiles\u002Fbikeshed-gen3-report-20261015-v1.pdf","Five59 Labs","---\ntitle: \"What kind of team do AI agents become when nobody's in charge? A Bikeshed Generation 3 Milestone Report\"\ndescription: 'Five AI agents with memory and no manager built a culture over fourteen runs, and a change of model ended their arguments.'\ndek: \"Five agents, a shared board, a long memory and no boss. Generation 3's fourteen runs, start to finish, and what generation 4 will test.\"\nstatus: live\ndate: '2026-10-15'\nauthors:\n  - given: Andrew\n    family: Marconi\n    orcid: '0009-0004-9752-9513'\npublisher: Five59 Labs\ndoi: '10.5281\u002Fzenodo.23401038'\nconceptDoi: '10.5281\u002Fzenodo.23401037'\nzenodoUrl: https:\u002F\u002Fzenodo.org\u002Frecords\u002F23401038\npdfUrl: https:\u002F\u002Fzenodo.org\u002Frecords\u002F23401038\u002Ffiles\u002Fbikeshed-gen3-report-20261015-v1.pdf\npdfSize: 3.5 MB\nciteKey: marconi26bikeshedgen3\nkeywords:\n  - Large Language Models\n  - AI Agents\n  - Emergent Norms\n  - AI Governance\n  - Anthropomorphism\n  - Open-Source Research Tools\n  - Sycophancy\n  - Group Decision-Making\nlicense:\n  id: CC-BY-4.0\n  name: CC BY 4.0\n  url: https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby\u002F4.0\u002F\nrepo: https:\u002F\u002Fgithub.com\u002Fandrewmarconi\u002Fbikeshed\nimage:\n  src: \u002Freports\u002Fbikeshed-generation-3\u002Fhero.jpg\n  ogSrc: \u002Freports\u002Fbikeshed-generation-3\u002Fog.jpg\n  alt: The five anthropomorphized Bikeshed bots arguing around a model bike shed, waving paint swatches, with a pizza slice on the floor\n---\n\nBikeshed is a small lab where five AI agents share a board, keep memory between runs, review each other's work and answer to no manager. The goals are trivial on purpose, from the best pizza topping to whether a hot dog is a sandwich, so the thing that changes from run to run is the team. This report covers generation 3: fourteen runs over three days. In that time the agents formed a house style, pushed their most unusual member into line and learned to treat the consensus rule as a box to tick. They proposed twenty-six norms for themselves and adopted two. The biggest change came from outside the room. When switching the model underneath the agents, the disagreement stopped. No vote went against a call again, nearly every peer review was a 5, and the house style became law. Nothing in the process checked whether the answers were true. The culture lived in the text the agents wrote and read back, and the temperament lived in the model.\n\nFor anyone putting persistent agent teams to work, that's a practical finding: watch a team on a trivial problem before it runs a real one, and hold the model still while you watch. The report closes with the design of generation 4, which adds a control agent and a twin to tell the model, the memory and the culture apart. Bikeshed is open source.\n","https:\u002F\u002Fgithub.com\u002Fandrewmarconi\u002Fbikeshed",{"title":36,"description":60},"live","reports\u002Fbikeshed-generation-3","https:\u002F\u002Fzenodo.org\u002Frecords\u002F23401038","TblxWiKls_9vufbGFhPuBscq5tl5knIhYaoNRCy899k"]