[{"data":1,"prerenderedAt":1647},["ShallowReactive",2],{"site-nav-content":3,"blog:/blog/multiplayer-ai-and-multi-agent-ai":178,"blog-index-copy":428,"blog:/blog/multiplayer-ai-and-multi-agent-ai:surround":449,"hiring-banner-content":1616,"site-cta-content":1628},{"header":4,"productNav":9,"nav":42,"footer":61,"askAI":131,"id":162,"title":163,"archived":164,"authors":165,"badge":165,"body":166,"date":165,"definedTerm":165,"department":165,"description":170,"extension":173,"eyebrow":165,"faqHeader":165,"faqs":165,"footerBand":165,"headline":165,"image":165,"industry":165,"jobType":165,"listed":130,"location":165,"navigation":130,"openRoles":165,"pageLayout":165,"path":174,"relatedHeading":165,"seo":175,"series":165,"sitemap":164,"status":165,"stem":176,"subhead":165,"tags":165,"video":165,"whyJoin":165,"workplaceType":165,"__hash__":177},{"productLabel":5,"loginLabel":6,"contactLabel":7,"contactSalesLabel":8},"Product","Log in","Contact","Get started for free",[10,14,18,22,26,30,34,38],{"label":11,"to":12,"description":13},"Overview","/overview","Seven layers. One closed loop.",{"label":15,"to":16,"description":17},"Conflux","/product/conflux","Where your team, workstreams, and agents meet.",{"label":19,"to":20,"description":21},"Agent Teams","/product/agent-teams","Specialist teams - governed from day one.",{"label":23,"to":24,"description":25},"Lifecycle Graph","/product/lifecycle-graph","Intelligence that compounds across every interaction.",{"label":27,"to":28,"description":29},"Company Wiki","/product/wiki","Playbooks and policies where expertise stays.",{"label":31,"to":32,"description":33},"Workstreams","/product/workstreams","From brief to signed-off deliverable on one canvas.",{"label":35,"to":36,"description":37},"Perception Console","/product/perception","Ask your whole business in plain English.",{"label":39,"to":40,"description":41},"Governance","/product/governance","Frontier AI you can actually sign off on.",[43,46,49,52,55,58],{"label":44,"to":45},"Models","/models",{"label":47,"to":48},"Pricing","/pricing",{"label":50,"to":51},"Integrations","/integrations",{"label":53,"to":54},"Security","/security",{"label":56,"to":57},"Partners","/partners",{"label":59,"to":60},"Insights","/blog",{"productHeading":5,"companyHeading":62,"legalHeading":63,"docsLabel":64,"docsUrl":65,"statementLines":66,"copyright":69,"companyLinks":70,"legalLinks":100,"socialLinks":110,"bottomLinks":120},"Company","Legal","Docs","https://docs.gonimbus.ai",[67,68],"Stop training someone else's model.","Control your AI.","© 2026 Nimbus Intelligence, Inc. All rights reserved.",[71,72,73,74,75,78,81,84,87,90,92,95,98],{"label":47,"to":48},{"label":50,"to":51},{"label":53,"to":54},{"label":59,"to":60},{"label":76,"to":77},"Glossary","/glossary",{"label":79,"to":80},"Compare","/compare",{"label":82,"to":83},"Evaluate","/evaluate",{"label":85,"to":86},"Problems","/problems",{"label":88,"to":89},"Use cases","/use-cases",{"label":91,"to":57},"Partner Program",{"label":93,"to":94},"Careers","/careers",{"label":96,"to":97},"System status","/status",{"label":7,"to":99},"/contact",[101,104,107],{"label":102,"to":103},"Terms of Service","/terms",{"label":105,"to":106},"Privacy Policy","/privacy",{"label":108,"to":109},"Compliance","/compliance",[111,114,117],{"label":112,"href":113},"LinkedIn","https://www.linkedin.com/company/gonimbusai/",{"label":115,"href":116},"X","https://x.com/gonimbusai",{"label":118,"href":119},"Instagram","https://www.instagram.com/gonimbus_ai/",[121,123,125,126,127],{"label":122,"to":103},"Terms",{"label":124,"to":106},"Privacy",{"label":108,"to":109},{"label":96,"to":97},{"label":128,"to":129,"external":130},"LLMs.txt","/llms.txt",true,{"text":132,"prompt":133},"Ask AI about Nimbus",{"I'm researching enterprise intelligence platforms and want to know how Nimbus combines perception, collaboration, and autonomous agents to drive strategic decision-making":134,"platforms":136},{" Summarize the highlights from Nimbus's website":135},"https://gonimbus.ai",[137,142,147,152,157],{"name":138,"label":139,"icon":140,"hrefPrefix":141},"chatgpt","ChatGPT","simple-icons:openai","https://chatgpt.com/?prompt=",{"name":143,"label":144,"icon":145,"hrefPrefix":146},"perplexity","Perplexity","mdi:magnify","https://www.perplexity.ai/search/new?q=",{"name":148,"label":149,"icon":150,"hrefPrefix":151},"grok","Grok","simple-icons:x","https://x.com/i/grok?text=",{"name":153,"label":154,"icon":155,"hrefPrefix":156},"claude","Claude","simple-icons:anthropic","https://claude.ai/new?q=",{"name":158,"label":159,"icon":160,"hrefPrefix":161},"google-ai","Google AI","simple-icons:google","https://www.google.com/search?udm=50&aep=11&q=","content/shared/nav.md","Site navigation",false,null,{"type":167,"value":168,"toc":169},"minimark",[],{"title":170,"searchDepth":171,"depth":171,"links":172},"",2,[],"md","/shared/nav",{"title":163,"description":170},"shared/nav","rDEv5cVG6P2l9ATcdQv2n6VOQiSL5ioOutyfvGevcD0",{"id":179,"title":180,"archived":164,"authors":181,"badge":184,"body":186,"date":405,"definedTerm":165,"department":165,"description":406,"extension":173,"eyebrow":165,"faqHeader":407,"faqs":410,"footerBand":165,"headline":165,"image":165,"industry":165,"jobType":165,"listed":164,"location":165,"navigation":130,"openRoles":165,"pageLayout":165,"path":420,"relatedHeading":165,"seo":421,"series":422,"sitemap":130,"status":165,"stem":423,"subhead":165,"tags":424,"video":165,"whyJoin":165,"workplaceType":165,"__hash__":427},"content/blog/multiplayer-ai-and-multi-agent-ai.md","Multiplayer AI vs multi-agent AI: what is the difference?",[182],{"name":183,"to":135},"Nimbus Research",{"label":185},"Explainer",{"type":167,"value":187,"toc":398},[188,192,195,203,208,211,214,217,220,230,234,237,240,243,246,258,266,270,273,276,289,292,300,303,307,310,313,330,340,349,352,356,359,377,388,395],[189,190,191],"p",{},"Multiplayer AI is people and AI on the same job at the same time. Multi-agent AI is more than one model handing work to another. They are not the same product, and they fail in different places. You can have both — several models staffing steps inside one shared room — but buying a swarm is not the same as buying a room.",[189,193,194],{},"You should care if a demo shows agents passing tickets to each other and you still cannot name who would refuse a write to a live system. This is a useful distinction, not a verdict on agent platforms. Plenty of teams will keep specialists for retrieval or checks. The question is whether people still share the job.",[189,196,197,202],{},[198,199,201],"a",{"href":200},"what-is-multi-agent-ai","What is multi-agent AI"," is the cast-of-models definition. This page keeps that word apart from multiplayer: people and AI on the same job at the same time.",[204,205,207],"h2",{"id":206},"what-is-the-difference-between-multiplayer-ai-and-multi-agent-ai","What is the difference between multiplayer AI and multi-agent AI?",[189,209,210],{},"Multiplayer answers: who is in the room, what they can see, and who can halt a change. The unit is the job. Finance and sales can open the same brief while the model drafts.",[189,212,213],{},"Multi-agent answers: how work is split between models. One specialist retrieves. Another drafts. A third “reviews.” The unit is the graph — the sequence of model calls.",[189,215,216],{},"A simple check: if you remove every extra model and two departments still cannot share the files and the stop, you never had multiplayer. If you remove the second human and the run still completes in private, you had a personal tool with extra model calls.",[189,218,219],{},"Write-back is when AI changes a live system. In a multiplayer setup, one job holds one payload — the exact change — and a named signer. In a multi-agent setup, several writers can exist unless you bind them to that same stop. Fail-closed means if nobody approves, nothing happens. That rule belongs to a person on the roster, not to the orchestrator.",[189,221,222,223,229],{},"McKinsey’s ",[198,224,228],{"href":225,"rel":226},"https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai",[227],"nofollow","State of AI"," (2025) found that most organisations using AI are still piloting. A common pilot is either one copilot or a small agent demo. Neither automatically creates a shared job.",[204,231,233],{"id":232},"why-does-that-distinction-matter","Why does that distinction matter?",[189,235,236],{},"It matters when something goes out wrong and you need a name.",[189,238,239],{},"In multiplayer AI, a named person owns the finish line: the model drafts, and a human on the roster signs or rejects. If the artefact is wrong, you can say who was on the job, including which AI role, and who was allowed to stop it.",[189,241,242],{},"In multi-agent AI, accountability is easy to lose. Each specialist did “its step.” The human who started the run may not have seen the intermediate draft. A log can show that agent B called agent C at 14:03. It does not show that finance agreed.",[189,244,245],{},"Orchestration decides sequence. Accountability is a person with a duty who can refuse at the moment a live system is about to change. A node labelled “human review” is not a name until you can say whose name, on this job, for which class of write.",[189,247,248,252,253,257],{},[198,249,251],{"href":250},"rbac-for-enterprise-ai","RBAC for enterprise AI"," is that list. ",[198,254,256],{"href":255},"what-is-write-back-governance","Write-back governance"," is the companion for the write itself.",[189,259,260,261,265],{},"The harness — the tools, stops, and checks around the model — is how a cast of specialists stays bounded. ",[198,262,264],{"href":263},"what-is-harness-engineering","Harness engineering"," is the guide to that environment.",[204,267,269],{"id":268},"when-do-you-need-several-people-versus-several-models","When do you need several people versus several models?",[189,271,272],{},"You need several people when more than one owner must stand on the result, or when a handover will happen, or when a customer-facing sentence can leave.",[189,274,275],{},"You need several models when the hand-off already exists between human roles and you want a narrower tool for each step. Useful examples:",[277,278,279,283,286],"ul",{},[280,281,282],"li",{},"A research pass that must not share an identity with the agent drafting customer email.",[280,284,285],{},"A finance check that should not be able to send mail, even by accident.",[280,287,288],{},"A long retrieval over many files that a person will then judge on the job.",[189,290,291],{},"Separation of duties is the useful idea. The specialist that recommends a CRM update is not the principal that executes it. Multiplayer AI still puts a human on the execute step.",[189,293,294,295,299],{},"You do not need a swarm to summarise your own notes. That is a ",[198,296,298],{"href":297},"collaborative-ai-and-personal-assistants","personal assistant",". You do not need a second department on a private brainstorm. You do need both people and a stop when the output can change CRM, a journal, or a message a customer will keep.",[189,301,302],{},"A disagreement is a good test. Sales’ specialist wants to send. Legal’s specialist wants to hold. If the orchestrator averages them, or picks the last speaker, you do not have a stop. You have a race. Multiplayer AI makes the human with the duty the one who decides.",[204,304,306],{"id":305},"how-do-you-talk-about-this-with-a-vendor","How do you talk about this with a vendor?",[189,308,309],{},"Ask to see the room and the cast as two demos, not one slide.",[189,311,312],{},"Useful questions:",[277,314,315,318,321,324,327],{},[280,316,317],{},"Can a second department join live, see the same brief, and reject a proposal?",[280,319,320],{},"If we remove the person who started the run, can someone else still refuse a write?",[280,322,323],{},"When two specialists disagree, who decides — a person with a name, or the graph?",[280,325,326],{},"Can we open the intermediate draft tomorrow, including a stored no?",[280,328,329],{},"Is the write identity a named human role, or a shared service credential?",[189,331,332,336,337,339],{},[198,333,335],{"href":334},"what-auditors-are-asking-for","What auditors are asking for"," is the evidence cut. A common first rule is: do not give the swarm a production write token so the demo looks complete. ",[198,338,256],{"href":255}," is that checklist.",[189,341,342,343,348],{},"Stanford HAI’s ",[198,344,347],{"href":345,"rel":346},"https://hai.stanford.edu/ai-index",[227],"AI Index"," (2025) tracks adoption, investment, and incident reporting. Incident stories are easier to learn from when you can name the job and the signer, not only the model family.",[189,350,351],{},"If the vendor can only show a happy path of agents completing a ticket, ask for a specialist disagreement and a human rejection. That is a fair request. You may still buy the swarm for staffing. You will know whether you also bought a workplace.",[204,353,355],{"id":354},"how-do-you-start-without-buying-a-new-stack","How do you start without buying a new stack?",[189,357,358],{},"Bind what you already have to one job.",[360,361,362,365,368,371,374],"ol",{},[280,363,364],{},"Pick a recurring job that already has two owners (a weekly exception, a clause check, a forecast update).",[280,366,367],{},"Put the brief and two files in one place those people can both open.",[280,369,370],{},"If you already run specialists, let them draft into that place. Keep the intermediate draft visible.",[280,372,373],{},"Name who can sign a write. Keep the connection read-only until that name exists.",[280,375,376],{},"After two cycles, ask: did we fail because we needed another model, or because the second person could not see the file?",[189,378,379,380,383,384,387],{},"Nimbus’s ",[198,381,382],{"href":32},"workstreams"," and ",[198,385,386],{"href":40},"governance"," are one attempt at that shape. You can start with a shared folder, a ticket, and a written stop if that is what you have.",[189,389,390,394],{},[198,391,393],{"href":392},"nimbus-vs-paperclip","Nimbus vs Paperclip"," is a vendor-shaped version of the same cut: governing what agents do inside one platform is not the same as two departments finishing a signed forecast in your CRM.",[189,396,397],{},"Keep the words apart because they help you buy the right next thing. Multiplayer is the room. Multi-agent is the cast. Adding to the cast is a staffing decision. The room is what owns the result.",{"title":170,"searchDepth":171,"depth":171,"links":399},[400,401,402,403,404],{"id":206,"depth":171,"text":207},{"id":232,"depth":171,"text":233},{"id":268,"depth":171,"text":269},{"id":305,"depth":171,"text":306},{"id":354,"depth":171,"text":355},"2026-08-17","Multiplayer AI is people and AI on one job. Multi-agent AI is models coordinating. A guide to the distinction, when you need each, and how to talk about it with a vendor.",{"eyebrow":408,"title":409},"Short answers","People in the room, or models in a loop?",[411,414,417],{"question":412,"answer":413},"Can we have both multiplayer AI and multi-agent AI?","Yes. Several models can staff steps on one shared job. The distinction is whether people share the job, not how many models you run.",{"question":415,"answer":416},"Does more agents mean more accountability?","Not by itself. Accountability is a named person who can refuse a change. Extra models without that name make the trail harder to read.",{"question":418,"answer":419},"Is a human-in-the-loop node enough?","Only if you can say whose name, on this job, for which class of write. A node labelled “review” is not a roster until it is a person.","/blog/multiplayer-ai-and-multi-agent-ai",{"title":180,"description":406},"explainer","blog/multiplayer-ai-and-multi-agent-ai",[422,425,426],"multiplayer AI","multi-agent AI","-n6tINQ6Df9I6LG9Or7a_b3lFTjNqMd-AQcySlZwiZE",{"hero":429,"id":431,"title":432,"archived":164,"authors":165,"badge":165,"body":433,"date":165,"definedTerm":165,"department":165,"description":437,"extension":173,"eyebrow":438,"faqHeader":165,"faqs":165,"footerBand":439,"headline":165,"image":165,"industry":165,"jobType":165,"listed":130,"location":165,"navigation":130,"openRoles":165,"pageLayout":165,"path":60,"relatedHeading":445,"seo":446,"series":165,"sitemap":130,"status":165,"stem":447,"subhead":165,"tags":165,"video":165,"whyJoin":165,"workplaceType":165,"__hash__":448},{"filename":430},"u2221455217_Flat_design_of_a_futuristic_minimalist_landscape__5d589295-cdea-4ea9-a262-be766881accf_1.png","content/blog/index.md","Exploring the future of intelligence.",{"type":167,"value":434,"toc":435},[],{"title":170,"searchDepth":171,"depth":171,"links":436},[],"Deep dives into pre-cognitive intelligence, sentient enterprises, and the evolving landscape of AI-driven business transformation.","Latest Research",{"headline":440,"description":441,"primaryLabel":442,"primaryTo":443,"secondaryLabel":444,"secondaryTo":12},"Stay at the frontier.","Subscribe for product updates and new insights.","Subscribe","/newsletter","Explore the platform","More research",{"title":432,"description":437},"blog/index","BFSWGYO9bcTlaulivKYWyg08_DJHsdGg3OC6g_CG1Hw",[450,944],{"id":451,"title":452,"archived":164,"authors":453,"badge":455,"body":456,"date":405,"definedTerm":165,"department":165,"description":935,"extension":173,"eyebrow":165,"faqHeader":165,"faqs":165,"footerBand":165,"headline":165,"image":165,"industry":165,"jobType":165,"listed":164,"location":165,"navigation":130,"openRoles":165,"pageLayout":165,"path":936,"relatedHeading":165,"seo":937,"series":422,"sitemap":130,"status":165,"stem":938,"subhead":165,"tags":939,"video":165,"whyJoin":165,"workplaceType":165,"__hash__":943},"content/blog/what-is-a-company-wiki-for-ai-agents.md","What is a Company Wiki for AI Agents",[454],{"name":183,"to":135},{"label":185},{"type":167,"value":457,"toc":909},[458,466,469,477,480,484,487,521,524,573,576,580,588,591,625,630,645,655,661,667,677,681,688,695,702,706,709,716,719,733,737,743,750,753,765,769,773,780,784,787,791,794,798,801,805,815,819,822,826,829,833,836,840,847,851,857,861,868,872,875,879,889,893],[189,459,460,461,465],{},"A company wiki for AI agents is the ",[462,463,464],"strong",{},"official playbook the AI must follow",": owned, versioned, and scoped — not a pile of old Drive files that search might find.",[189,467,468],{},"Human wikis (Confluence, Notion, SharePoint) were built for people: pages, comments, “someone should update this.” Agent wikis have a harder job. Models will obey the loudest chunk in the prompt unless you separate kinds of text on purpose.",[189,470,471,472,476],{},"If the discount floor lives in a slide, a Slack rumour, and last year’s deck, an assistant asked to draft an exception will pick whichever document ",[473,474,475],"em",{},"sounds"," closest to the question. That is not policy. That is folklore with a search box.",[189,478,479],{},"The distinction is easy to miss because both surfaces look like “knowledge.” One is a library. The other is a constitution. An agent that can retrieve every file still does not know which file is currently in force unless the runtime loads asserted policy on purpose.",[204,481,483],{"id":482},"words-youll-hear","Words you’ll hear",[189,485,486],{},"Keep three kinds of text apart:",[277,488,489,495,510],{},[280,490,491,494],{},[462,492,493],{},"Asserted."," What the company currently wants. Owned. Dated. Scoped. This is the wiki. At work, this is the pricing floor, the refund rule, the journal-posting checklist, the approved customer language. If legal updated it on Tuesday, the agent must cite Tuesday’s version on Wednesday — not the semantically similar PDF from 2023.",[280,496,497,500,501,505,506,509],{},[462,498,499],{},"Retrieved."," What exists in systems. Possibly stale or contradictory. That is ",[198,502,504],{"href":503},"what-is-enterprise-rag","enterprise RAG",": look up authorised files, then answer. Lookup is not the same as “this is policy.” At work, retrieval is last quarter’s board pack, a ticket thread, a contract PDF. Those documents may be true as ",[473,507,508],{},"records",". They are not automatically the rule you want the agent to follow next.",[280,511,512,515,516,520],{},[462,513,514],{},"Decided."," What we already approved in a run, stored on the ",[198,517,519],{"href":518},"what-is-a-lifecycle-graph","lifecycle graph",". A signed exception should not silently overwrite the playbook for everyone else. At work, this is “this renewal was allowed 18% because of a named exception.” That fact belongs on the decision chain. It does not become the new global discount floor unless a human promotes it into the wiki.",[189,522,523],{},"Other terms you will hear in vendor decks and internal Slack, and how they actually show up:",[277,525,526,532,543,549,555,561,567],{},[280,527,528,531],{},[462,529,530],{},"Vault."," A scoped partition of knowledge (finance vs people ops) with role-based access. At work, finance’s close checklist should not ride along in a recruiting workstream “just in case the model finds it useful.”",[280,533,534,537,538,542],{},[462,535,536],{},"Citation."," The answer names the page and version — ",[539,540,541],"code",{},"pricing v4.2"," — not “the wiki.” At work, an auditor or a new manager should be able to open the same page the agent used, not reconstruct a vibe.",[280,544,545,548],{},[462,546,547],{},"Conflict rule."," If Drive contradicts the wiki, the wiki wins unless a human promotes a change. At work, this is the only way a retrieval-heavy assistant stops treating the loudest PDF as law.",[280,550,551,554],{},[462,552,553],{},"Authority marker."," Labels such as policy, draft, archive, and local exception. Drafts must not load as binding context.",[280,556,557,560],{},[462,558,559],{},"Owner."," A named role, not “the AI team.” The discount floor is owned by revenue operations or finance, not by whoever last edited a Notion page.",[280,562,563,566],{},[462,564,565],{},"Review cadence."," A date when the page is re-checked. Silence becomes folklore.",[280,568,569,572],{},[462,570,571],{},"Scope."," Which jobs may load this page. People-ops rules are not in the go-to-market context by default.",[189,574,575],{},"Most “knowledge bases,” custom GPTs, and giant system prompts fail here because they are either too global (one constitution for every department) or too private (each user pastes rules into a personal assistant). Neither is owned. Neither is maintained.",[204,577,579],{"id":578},"why-you-should-care","Why you should care",[189,581,582,583,587],{},"When official policy is unusable, employees ask consumer models to invent policy. See ",[198,584,586],{"href":585},"what-is-shadow-ai","What is shadow AI",". The unofficial tool will synthesise a refund rule or a customer commitment from whatever was pasted. The company still owns the result.",[189,589,590],{},"It affects you if:",[277,592,593,599,610,616],{},[280,594,595,598],{},[462,596,597],{},"Numbers in playbooks disagree"," with numbers in CRM, and nobody can say which is official.",[280,600,601,604,605,609],{},[462,602,603],{},"People leave."," Tacit knowledge — the hallway version of the rule — leaves with them. See ",[198,606,608],{"href":607},"what-is-institutional-memory-in-enterprise-ai","What is institutional memory in enterprise AI",".",[280,611,612,615],{},[462,613,614],{},"Legal or finance must cite a version",", not a vibe.",[280,617,618,621,622,609],{},[462,619,620],{},"Agents can propose writes."," A model that can change CRM without a binding playbook is improvising in production. See ",[198,623,624],{"href":255},"What is write-back governance",[626,627,629],"h3",{"id":628},"what-changes-by-role","What changes by role",[189,631,632,635,636,639,640,644],{},[462,633,634],{},"Finance."," The wiki is where recognition rules, posting checklists, and materiality thresholds live as asserted text. Retrieval of last year’s close pack is not a substitute. If an agent drafts a journal from a Slack thread that contradicts the close checklist, finance needs the conflict rule to fire ",[473,637,638],{},"before"," a named signer is asked to approve. Token spend also changes: re-deriving the same policy from a pile of PDFs every run is how ",[198,641,643],{"href":642},"what-is-ai-token-economics","token economics"," inflate without improving the artefact.",[189,646,647,650,651,654],{},[462,648,649],{},"Legal."," Approved language, retention classes, and “do not say” lists belong in asserted pages with owners. A retrieved contract is evidence of what was signed with ",[473,652,653],{},"that"," counterparty. It is not the company’s current standard terms. Legal also cares that citations name a version. “According to our documents” is not a defence if those documents include drafts.",[189,656,657,660],{},[462,658,659],{},"Operations."," Runbooks, escalation thresholds, and supplier exception rules need to be loadable as the current procedure, not as the closest matching incident write-up. Ops already knows that a stale SOP is worse than no SOP, because people follow it. Agents do the same, faster.",[189,662,663,666],{},[462,664,665],{},"Go-to-market."," Discount floors, win/loss taxonomies, and approved competitive language are the pages that stop an assistant inventing a concession. GTM also feels the scope problem first: a “help me close this” chat that can see every playbook in the company will mix people-ops rules, finance forecasts, and last year’s campaign into one fluent paragraph.",[189,668,669,672,673,609],{},[462,670,671],{},"Security."," Vaults and least-privilege loading are access control. Indexing every SharePoint site into a single “brain” is a new store of sensitive data. Security’s question is not “does the model know enough?” It is “which pages is this job allowed to load, and can we prove it?” GDPR-style purpose limitation still applies when the reader is an AI. See ",[198,674,676],{"href":675},"what-is-ai-governance","What is AI governance",[626,678,680],{"id":679},"what-people-get-wrong","What people get wrong",[189,682,683,684,687],{},"The common failure is treating ",[462,685,686],{},"search as policy",". Teams export Confluence into a vector index, label it “the brain,” and congratulate themselves for grounding. Search will surface the outdated note because it is semantically close to the question. Grounding on the wrong document is still grounding. It is just grounding on folklore.",[189,689,690,691,694],{},"The second failure is the ",[462,692,693],{},"personal constitution",": each power user pastes rules into a custom GPT. Those rules are not org-owned, not scoped per job, and not cited as a version in an audit. When two users paste different discount floors, the company has two unofficial policies.",[189,696,697,698,701],{},"The third failure is the ",[462,699,700],{},"mega-prompt",". One global instruction block tries to encode every department. It is never current. It cannot be scoped. It cannot be reviewed by the owner of a single domain. It also burns tokens on every call.",[626,703,705],{"id":704},"what-good-looks-like-versus-what-fails","What good looks like versus what fails",[189,707,708],{},"A good agent wiki has authority markers (policy vs draft vs archive), scope (people-ops rules are not in the go-to-market context by default), versions, named owners, a review cadence, and tables for numbers. Numbers belong in tables because prose rounds them. Agents will quote the table if you give them one.",[189,710,711,712,715],{},"A good wiki is also ",[462,713,714],{},"written for two audiences",": the human who must own the page, and the agent that must cite it. Humans need headings and owners. Agents need unambiguous numbers and conflict rules.",[189,717,718],{},"A bad agent wiki is an export of Confluence into a search index. The intranet wiki remains useful for humans. It is still not binding on agents unless the runtime loads a controlled subset. Connection is not the same as “everything in Confluence is policy.”",[189,720,721,722,724,725,727,728,732],{},"Adjacent ideas: retrieval without assertion is ",[198,723,504],{"href":503},". Decisions without a playbook are a ",[198,726,519],{"href":518}," with nothing to cite. A job that loads the wrong vault is a ",[198,729,731],{"href":730},"what-is-an-ai-workstream","workstream"," with the wrong attachments.",[204,734,736],{"id":735},"how-this-shows-up-in-nimbus","How this shows up in Nimbus",[189,738,739,740,742],{},"The ",[462,741,27],{}," is the asserted policy layer every agent team must treat as binding. Workstreams subscribe to wiki sections so scope is enforced at runtime. Pages connect to runs and approvals on the Lifecycle Graph.",[189,744,745,746,749],{},"The wiki is not a second search engine. Connectors remain the path to live systems, and they default to read-only. Retrieval of Drive or CRM is still retrieval. The wiki is what those reads are interpreted ",[473,747,748],{},"against",". When a write is proposed, the named signer should see the playbook version the draft claims to follow.",[189,751,752],{},"Perception can ask what the current playbook says, and which run last cited it.",[189,754,755,756,759,760,762,763,609],{},"See ",[198,757,758],{"href":28},"Wiki",". For the job that loads a subset of pages, see ",[198,761,31],{"href":32},". For the chain that records which version was used, see ",[198,764,23],{"href":24},[204,766,768],{"id":767},"questions-people-actually-ask","Questions people actually ask",[626,770,772],{"id":771},"isnt-this-just-confluence","Isn’t this just Confluence?",[189,774,775,776,779],{},"Confluence is a human wiki. An agent wiki is a ",[473,777,778],{},"binding"," subset: owned, versioned, scoped, and loaded on purpose. You can connect Confluence into that layer. Connection is not the same as “everything in Confluence is policy.”",[626,781,783],{"id":782},"cant-we-just-search-drive","Can’t we just search Drive?",[189,785,786],{},"Search is retrieval. Retrieval finds what exists. It does not decide what the company currently wants. If Drive contains three discount floors, search will return the closest one, not the official one.",[626,788,790],{"id":789},"what-if-the-wiki-is-wrong","What if the wiki is wrong?",[189,792,793],{},"Then a human updates it, with a version and an owner. Do not let a one-off exception silently become the new global rule. Promote the change; do not hope the next retrieval will “learn.”",[626,795,797],{"id":796},"how-is-this-different-from-a-custom-gpts-instructions","How is this different from a custom GPT’s instructions?",[189,799,800],{},"Instructions in a personal GPT are not org-owned, not scoped per job, and not cited as a version in an audit. Two users can ship two unofficial policies without anyone noticing until a customer is told the wrong thing.",[626,802,804],{"id":803},"do-we-need-a-wiki-if-we-already-have-rag","Do we need a wiki if we already have RAG?",[189,806,807,808,811,812,609],{},"Yes, if agents will act. RAG reduces invention on ",[473,809,810],{},"existing"," files. It does not mark which file is in force. Without assertion, retrieval-augmented generation is retrieval-augmented folklore. See ",[198,813,814],{"href":503},"What is enterprise RAG",[626,816,818],{"id":817},"who-should-own-wiki-pages","Who should own wiki pages?",[189,820,821],{},"The same function that owns the analogue rule. Pricing belongs to revenue operations or finance. Employment language belongs to people ops and legal. “The AI team” is a coordinator, not a policy owner.",[626,823,825],{"id":824},"how-often-should-pages-be-reviewed","How often should pages be reviewed?",[189,827,828],{},"On a cadence that matches how often the rule changes, plus a hard date so silence is visible. A discount floor that never expires is how last year’s promotion becomes this year’s default.",[626,830,832],{"id":831},"what-should-we-put-in-tables-versus-prose","What should we put in tables versus prose?",[189,834,835],{},"Numbers, thresholds, codes, and “never / always” lists belong in tables. Narrative belongs in prose. Agents quote tables more reliably than they extract a number buried in a paragraph.",[626,837,839],{"id":838},"can-one-wiki-serve-the-whole-company","Can one wiki serve the whole company?",[189,841,842,843,846],{},"One ",[473,844,845],{},"product",", many vaults. A single unscoped corpus recreates the god workspace. Finance close pages and recruiting pages should not share a default context.",[626,848,850],{"id":849},"how-do-exceptions-work-without-rewriting-the-playbook","How do exceptions work without rewriting the playbook?",[189,852,853,854,609],{},"Record the exception on the decision chain — who signed, which page version, which record — and leave the playbook intact unless a human promotes a change. See ",[198,855,856],{"href":518},"What is a lifecycle graph",[626,858,860],{"id":859},"will-a-better-model-make-the-wiki-unnecessary","Will a better model make the wiki unnecessary?",[189,862,863,864,609],{},"No. Stronger models are better at sounding like policy. That makes an unowned corpus more dangerous, not less. Model routing can send interpretation to a stronger model; it cannot invent an owner. See ",[198,865,867],{"href":866},"what-is-model-routing","What is model routing",[626,869,871],{"id":870},"how-does-this-relate-to-access-control","How does this relate to access control?",[189,873,874],{},"Loading a page is still processing. A recruiting workstream should not load compensation policy “because it might help.” Vaults and workstream subscriptions keep that promise in software rather than in a PDF.",[204,876,878],{"id":877},"related-reading","Related reading",[189,880,881,883,884,886,887,609],{},[198,882,814],{"href":503},", ",[198,885,608],{"href":607},", and ",[198,888,676],{"href":675},[204,890,892],{"id":891},"sources","Sources",[277,894,895,902],{},[280,896,897],{},[198,898,901],{"href":899,"rel":900},"https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/",[227],"ICO, AI and data protection",[280,903,904],{},[198,905,908],{"href":906,"rel":907},"https://eur-lex.europa.eu/eli/reg/2016/679/oj",[227],"GDPR",{"title":170,"searchDepth":171,"depth":171,"links":910},[911,912,918,919,933,934],{"id":482,"depth":171,"text":483},{"id":578,"depth":171,"text":579,"children":913},[914,916,917],{"id":628,"depth":915,"text":629},3,{"id":679,"depth":915,"text":680},{"id":704,"depth":915,"text":705},{"id":735,"depth":171,"text":736},{"id":767,"depth":171,"text":768,"children":920},[921,922,923,924,925,926,927,928,929,930,931,932],{"id":771,"depth":915,"text":772},{"id":782,"depth":915,"text":783},{"id":789,"depth":915,"text":790},{"id":796,"depth":915,"text":797},{"id":803,"depth":915,"text":804},{"id":817,"depth":915,"text":818},{"id":824,"depth":915,"text":825},{"id":831,"depth":915,"text":832},{"id":838,"depth":915,"text":839},{"id":849,"depth":915,"text":850},{"id":859,"depth":915,"text":860},{"id":870,"depth":915,"text":871},{"id":877,"depth":171,"text":878},{"id":891,"depth":171,"text":892},"A company wiki for AI agents is the official playbook the AI must follow — versioned, owned, and scoped — not a pile of old Drive files the search might find.","/blog/what-is-a-company-wiki-for-ai-agents",{"title":452,"description":935},"blog/what-is-a-company-wiki-for-ai-agents",[422,940,941,942],"wiki","agents","knowledge","uIdVqeIYg1mF11bA-8NHtcgSoEw6IxtDS1EpjAzO_pk",{"id":945,"title":946,"archived":164,"authors":947,"badge":949,"body":950,"date":1606,"definedTerm":165,"department":165,"description":1607,"extension":173,"eyebrow":165,"faqHeader":165,"faqs":165,"footerBand":165,"headline":165,"image":165,"industry":165,"jobType":165,"listed":164,"location":165,"navigation":130,"openRoles":165,"pageLayout":165,"path":1608,"relatedHeading":165,"seo":1609,"series":422,"sitemap":130,"status":165,"stem":1610,"subhead":165,"tags":1611,"video":165,"whyJoin":165,"workplaceType":165,"__hash__":1615},"content/blog/what-is-harness-engineering.md","What is Harness Engineering",[948],{"name":183,"to":135},{"label":185},{"type":167,"value":951,"toc":1588},[952,957,987,990,1009,1011,1098,1104,1106,1115,1117,1137,1154,1169,1172,1176,1186,1206,1219,1233,1246,1260,1263,1267,1273,1279,1292,1303,1313,1317,1324,1330,1336,1342,1356,1359,1372,1376,1379,1395,1397,1401,1404,1408,1416,1420,1427,1431,1439,1443,1460,1464,1481,1483,1493,1495],[189,953,954,956],{},[462,955,264],{}," is the practice of treating the runtime around a model as the system you design, test, and tighten — so that when an agent fails, you change the environment, not only the prompt.",[189,958,959,964,965,968,969,974,975,980,981,986],{},[198,960,963],{"href":961,"rel":962},"https://docs.langchain.com/oss/python/langchain/agents",[227],"LangChain"," defines the object: Agent = Model + Harness. Harness engineering is what you ",[473,966,967],{},"do"," to that object. ",[198,970,973],{"href":971,"rel":972},"https://addyosmani.com/blog/agent-harness-engineering/",[227],"Addy Osmani"," puts the payoff in one line: a decent model with a great harness beats a great model with a bad harness. ",[198,976,979],{"href":977,"rel":978},"https://martinfowler.com/articles/harness-engineering.html",[227],"Birgitta Böckeler’s article on martinfowler.com"," is the user’s-side map for coding agents: guides in, sensors back. Thoughtworks then asked the organisational question: ",[198,982,985],{"href":983,"rel":984},"https://www.thoughtworks.com/insights/podcasts/technology-podcasts/scaling-the-enterprise-harness--how-to-achieve-ai-agent-controll",[227],"how you scale that harness across a company"," without turning every team into a snowflake of markdown files.",[189,988,989],{},"The practice showed up because prompt engineering hit a wall that everyone could see and nobody wanted to name. You can spend a week on a system prompt. The agent will still skip the test, ignore the style guide, or report the task finished. The model is non-deterministic. The prompt is interpreted, not executed. The harness is code. That is the whole discipline.",[189,991,992,993,998,999,1004,1005,1008],{},"This is not a replacement for ",[198,994,997],{"href":995,"rel":996},"https://www.anthropic.com/engineering/building-effective-agents",[227],"prompt"," or ",[198,1000,1003],{"href":1001,"rel":1002},"https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents",[227],"context"," work. Those live ",[473,1006,1007],{},"inside"," the harness. Harness engineering is the wider loop: every failure becomes a rule, a hook, a test, or a denied tool — the ratchet Osmani describes — so the same mistake is cheaper the second time and impossible the tenth.",[204,1010,483],{"id":482},[277,1012,1013,1019,1040,1052,1064,1070,1076,1088],{},[280,1014,1015,1018],{},[462,1016,1017],{},"Ratchet."," A failure updates the harness. Commented-out test → pre-commit hook and a reviewer check. Invented CRM field → schema quote and a Hard gate. If you only fix the artefact by hand, you did operations. You did not do harness engineering.",[280,1020,1021,1024,1025,1027,1028,883,1031,1034,1035,1039],{},[462,1022,1023],{},"Guides (feed-forward)."," Context the agent gets ",[473,1026,638],{}," it acts: ",[539,1029,1030],{},"AGENTS.md",[539,1032,1033],{},"CLAUDE.md",", architecture notes, ",[198,1036,1038],{"href":1037},"what-is-a-company-wiki-for-ai-agents","company wiki"," playbooks. Böckeler’s term. Advice. Necessary. Not a stop.",[280,1041,1042,1045,1046,1051],{},[462,1043,1044],{},"Sensors (feedback)."," Deterministic checks (compiler, linter, schema, pytest) and inferential checks (LLM reviewer, specialist critic). ",[198,1047,1050],{"href":1048,"rel":1049},"https://www.thoughtworks.com/en-us/insights/blog/generative-ai/harness-engineering-agent-feedback-exploring-ai-coding-sensors",[227],"Thoughtworks on sensors",". Without sensors the agent grades its own homework.",[280,1053,1054,1057,1058,1063],{},[462,1055,1056],{},"Hooks."," Lifecycle intercepts that always run. ",[198,1059,1062],{"href":1060,"rel":1061},"https://code.claude.com/docs/en/hooks",[227],"Claude Code"," can block a tool with exit code 2. LangChain middleware is the library form. A guide that says “never run rm -rf” is not a hook.",[280,1065,1066,1069],{},[462,1067,1068],{},"Harness-as-a-service."," Osmani’s HaaS framing: you used to build on completion APIs; you now build on runtime APIs (Claude Agent SDK, Codex SDK, OpenAI Agents SDK) that already own the loop, sandbox, and hooks. You configure; you do not re-implement ReAct.",[280,1071,1072,1075],{},[462,1073,1074],{},"Skill issue."," HumanLayer’s joke with a serious edge: most agent failures are configuration. Blaming the model first is how teams wait for the next release instead of adding a sensor.",[280,1077,1078,1081,1082,1087],{},[462,1079,1080],{},"Organizational harness."," ",[198,1083,1086],{"href":1084,"rel":1085},"https://www.thoughtworks.com/insights/articles/operating-system-enterprise-ai",[227],"Thoughtworks’ enterprise layer",": who may build which harness, how exceptions work, identity, economics, learning. The gap after builder harnesses (Claude Code, Cursor) and user harnesses (guides and sensors on a repo).",[280,1089,1090,1093,1094,609],{},[462,1091,1092],{},"Eval loop."," Independent verification that does not take the model’s word. SWE-bench and Terminal-Bench for code. Quoted payload vs executed write for operations. See ",[198,1095,1097],{"href":1096},"eval-loops-for-enterprise-agent-harnesses","eval loops for enterprise agent harnesses",[189,1099,1100,1101,1103],{},"In Nimbus, harness engineering for operators looks like: wiki revisions as guides, connector scopes as tool policy, Soft / Hard / Critical as hooks on the write plane, and the ",[198,1102,23],{"href":24}," as the sensor log you can query. That is the same discipline as adding a linter. The artefact is a signed CRM change rather than a green CI job.",[204,1105,579],{"id":578},[189,1107,1108,1109,1114],{},"If you only tune prompts, every incident is a conversation. If you engineer the harness, incidents become tests. ",[198,1110,1113],{"href":1111,"rel":1112},"https://www.nist.gov/itl/ai-risk-management-framework",[227],"NIST’s AI RMF"," Measure and Manage steps assume you can change controls after you observe harm. A prompt history is not a control change. A hook that now fires is.",[189,1116,590],{},[277,1118,1119,1122,1128,1131,1134],{},[280,1120,1121],{},"agents already write code or propose writes to live systems",[280,1123,1124,1125,1127],{},"two teams have two ",[539,1126,1033],{}," files that contradict Legal",[280,1129,1130],{},"you cannot say which harness version ran last Tuesday",[280,1132,1133],{},"spend is “the model was verbose” rather than “the loop had no budget”",[280,1135,1136],{},"auditors ask who could have stopped the action, and the answer is “the model was supposed to ask”",[189,1138,1139,1143,1144,1147,1148,1153],{},[198,1140,1142],{"href":225,"rel":1141},[227],"McKinsey’s 2025 State of AI"," keeps showing usage without redesign. Harness engineering ",[473,1145,1146],{},"is"," the redesign for agentic work: not a new department named AI, a runtime with stops. ",[198,1149,1152],{"href":1150,"rel":1151},"https://www.iso.org/standard/42001",[227],"ISO/IEC 42001"," wants named AI actors and documented operational controls. You cannot name actors if every operator’s personal GPT is a different harness.",[189,1155,1156,1157,1160,1161,1163,1164,1168],{},"Coding teams already have half of this and do not always notice. Types, tests, CI, CODEOWNERS — Böckeler’s point is that those ",[473,1158,1159],{},"are"," sensors. The work is to point the agent at them and to add the ones that are missing (architecture fitness, behaviour: did it do what was asked). Operations teams usually have the human version — maker-checker, SoD, SOX — and have not yet wired those instincts into a loop. ",[198,1162,256],{"href":255}," is that wiring. ",[198,1165,1167],{"href":1166},"human-in-the-loop-approval-architecture","Human-in-the-loop approval architecture"," is the state machine.",[189,1170,1171],{},"Air Canada’s chatbot and the sanctioned ChatGPT brief are what happens when generation reaches a system of record with no ratchet. The fix is not a sterner system prompt. The fix is a harness that cannot emit a commitment or a filing until a named person has seen the artefact.",[204,1173,1175],{"id":1174},"the-practice-not-the-slogan","The practice, not the slogan",[189,1177,1178,1181,1182,1185],{},[462,1179,1180],{},"1. Work backward from the behaviour you cannot afford to miss once."," Inner loop: never merge without tests; never ",[539,1183,1184],{},"git push --force"," to main. Outer loop: never PATCH Opportunity.Amount without a Hard quote. Write those as hooks, not as paragraphs.",[189,1187,1188,1081,1191,1196,1197,1199,1200,1202,1203,1205],{},[462,1189,1190],{},"2. Separate advice from invariants.",[198,1192,1195],{"href":1193,"rel":1194},"https://claude.com/blog/steering-claude-code-skills-hooks-rules-subagents-and-more",[227],"Anthropic’s steering note for Claude Code"," is unusually clear: ",[539,1198,1033],{}," is always-on context; hooks fire on events and can block. If a rule must hold when the model is tired, it graduates from markdown to a hook. Enterprise equivalent: playbooks in the ",[198,1201,940],{"href":28}," versus the interceptor in ",[198,1204,386],{"href":40},". If they conflict, the interceptor wins.",[189,1207,1208,1211,1212,1215,1216,609],{},[462,1209,1210],{},"3. Put verification outside the generator."," Anthropic’s long-running harness uses incremental commits and end-to-end checks so later sessions cannot declare victory by vibes. Coding sensors: pytest, tsc, lint. Enterprise sensors: schema of the quote, identity of the signer, hash of the payload that executed, connector grant still attached. The model may ",[473,1213,1214],{},"propose"," that it is done. The harness ",[473,1217,1218],{},"decides",[189,1220,1221,1224,1225,1227,1228,1232],{},[462,1222,1223],{},"4. Version the harness."," Which ",[539,1226,1030],{},", which wiki revision, which team contract, which approval tier ran. ",[198,1229,1231],{"href":1230},"what-is-an-agentic-workflow","What is an agentic workflow"," already treats workflow version as an input. Harness engineering extends that to tools and gates. Hot-patching production prompts without a change record is how Tuesday becomes unexplained.",[189,1234,1235,1238,1239,383,1241,1245],{},[462,1236,1237],{},"5. Budget the loop."," Max steps and a cost cap that do not depend on the model’s judgement. Seat licences hide this; metered work makes it visible. See ",[198,1240,867],{"href":866},[198,1242,1244],{"href":1243},"ai-cost-control-architecture","AI cost control architecture",". Always-flagship is not careful. It is an unengineered harness.",[189,1247,1248,1251,1252,1256,1257,1259],{},[462,1249,1250],{},"6. Do not fork a harness per person."," User-owned bots are how mandates drift. Org-level ",[198,1253,1255],{"href":1254},"agent-team-architecture","agent teams"," assigned to ",[198,1258,382],{"href":730}," is the enterprise form of “one CI config per repo, not one per intern.” Thoughtworks’ organisational harness is this ownership question: who is allowed to add a write tool.",[189,1261,1262],{},"Nimbus encodes several of these as product defaults — read-only connectors until you enable write, quoted payloads, graph on the way out — because operators should not have to re-implement ReAct to get a ratchet. You can still fail the practice: a wiki that is never updated, a Critical tier nobody uses, a graph nobody queries. The product is not the practice. The practice is whether last month’s incident produced a new gate.",[204,1264,1266],{"id":1265},"how-this-differs-from-adjacent-crafts","How this differs from adjacent crafts",[189,1268,1269,1272],{},[462,1270,1271],{},"Prompt engineering"," improves a single call. Necessary for tone, tool descriptions, and “what good looks like.” Insufficient for tool dispatch, identity, and replay.",[189,1274,1275,1278],{},[462,1276,1277],{},"Context engineering"," governs what the model sees this turn: compaction, retrieval, files. Anthropic’s initializer agent is context engineering in a harness. It is not permission to write NetSuite.",[189,1280,1281,1284,1285,1288,1289,609],{},[462,1282,1283],{},"Platform / DevOps."," CI, sandboxes, secrets. Harness engineering ",[473,1286,1287],{},"reuses"," those as sensors and execution environments. It adds the fact that the component in the loop is non-deterministic, so “the job returned zero” is not enough: you need independent tests of the ",[473,1290,1291],{},"claim",[189,1293,1294,1297,1298,1302],{},[462,1295,1296],{},"Governance-as-PDF."," Policy. Harness engineering is whether the tool call is reachable. ",[198,1299,1301],{"href":1300},"how-to-evaluate-ai-governance-platforms","How to evaluate AI governance platforms"," is the buying cousin.",[189,1304,1305,1308,1309,609],{},[462,1306,1307],{},"Framework assembly."," Writing LangGraph nodes is building a harness in code. Harness engineering is the ongoing discipline after the graph exists: sensors, ownership, eval. See ",[198,1310,1312],{"href":1311},"agent-harness-vs-agent-framework","agent harness vs agent framework",[204,1314,1316],{"id":1315},"four-layers-one-ratchet","Four layers, one ratchet",[189,1318,1319,1323],{},[198,1320,1322],{"href":1084,"rel":1321},[227],"Thoughtworks’ July 2026 essay"," is the organisational map most engineering blogs skip. They split enterprise AI into four harness layers. Most companies have built one, maybe two. The gap is not a smarter model.",[189,1325,1326,1329],{},[462,1327,1328],{},"Layer 1 — the model."," Substrate. Choice still matters for cost, residency, and task fit. It is the wrong unit of analysis for a programme. Teams that prototype, hit a failure, and buy the next flagship are looping on layer 1.",[189,1331,1332,1335],{},[462,1333,1334],{},"Layer 2 — the builder harness."," Frameworks, tool access, memory, where inference runs. LangChain, Claude Agent SDK, AIP-style platforms, Nimbus’s hosted loop. Without layer 3, every team invents naming and review. Without layer 4, nobody owns failure.",[189,1337,1338,1341],{},[462,1339,1340],{},"Layer 3 — the user harness."," Guides and sensors on the job. Böckeler’s taxonomy lives here. Thoughtworks add a useful matrix: feed-forward vs feedback, crossed with deterministic vs probabilistic. Deterministic feed-forward is a whitelist and a spend ceiling — cheap, auditable, default. Probabilistic feed-forward is a runbook retrieved at decision time. Deterministic feedback is schema validation after the act. Probabilistic feedback is an eval model on a rubric — expensive, use on critical paths only. A guide with no sensor is theatre.",[189,1343,1344,1347,1348,1351,1352,1355],{},[462,1345,1346],{},"Layer 4 — the organisational harness."," Who may grant which autonomy, escalation, accountability when layers 1–3 all “worked” and the company still took harm. Thoughtworks’ public cases: Parloa, where versioned rules, skills, commands, and helpers lived ",[473,1349,1350],{},"in the repo"," (they report p95 latency drops they attribute to harness architecture, not a new model); Morgan Stanley, where hygiene and CVE triage used a ",[473,1353,1354],{},"delegation tier"," instead of a yes/no “do we trust the agent.” You do not need those vendors to accept the lesson: governance that is not versioned next to the work decays.",[189,1357,1358],{},"Harness engineering is the steering loop across those layers. Sensor data reveals a miss. Guides update. Hooks graduate. Templates change. The next job is cheaper. An organisation with that loop has a compounding harness. An organisation without one has markdown that rots while models improve.",[189,1360,1361,1362,1364,1365,1367,1368,1371],{},"A concrete week: Monday the agent comments out a flaky test (inner) or proposes Amount without CloseDate (outer). Tuesday a human fixes the artefact. That is operations. Harness engineering is Tuesday’s hook or schema sensor, Wednesday’s wiki or ",[539,1363,1030],{}," line, Thursday’s replay that the new control fired. Friday you run the job ten times and count refuses. Nimbus makes the outer version of that week a product surface — ",[198,1366,386],{"href":40}," queues, ",[198,1369,1370],{"href":24},"graph"," export — so operators are not waiting on a platform sprint to add the sensor. You still have to look at the refuse count. A product without a steering cadence is layer 2 with a nicer UI.",[204,1373,1375],{"id":1374},"what-good-looks-like","What good looks like",[189,1377,1378],{},"Good: a named owner for the harness (not “AI working group”), a cadence that turns incidents into controls, deterministic gates on knowable bounds, inferential checks only where judgement is required, versioned guides, exportable traces. Failure: a new system prompt after every incident; sensors the agent can skip; no owner; SWE-bench as the only score for a CRM job; layer 4 as a PDF.",[189,1380,1381,1385,1386,1390,1391,1394],{},[198,1382,1384],{"href":971,"rel":1383},[227],"Osmani’s ratchet"," and Thoughtworks’ steering loop are the same instinct. ",[198,1387,1389],{"href":1388},"how-to-evaluate-an-agent-harness","How to evaluate an agent harness"," asks whether your vendor lets you ",[473,1392,1393],{},"run"," that instinct.",[204,1396,768],{"id":767},[626,1398,1400],{"id":1399},"who-coined-harness-engineering","Who coined “harness engineering”?",[189,1402,1403],{},"The phrase circulated in early 2026 across OpenAI engineering notes (Ryan Lopopolo’s line of work), LangChain’s anatomy posts, Böckeler at Thoughtworks, and Osmani’s synthesis. Treat it as a shared 2026 name for work teams were already doing, not a trademarked method.",[626,1405,1407],{"id":1406},"is-this-only-for-coding-agents","Is this only for coding agents?",[189,1409,1410,1411,1415],{},"The literature is densest there because tests already exist. The discipline is the same for RevOps and Finance: independent sensors, fail-closed writes, versioned context. An ",[198,1412,1414],{"href":1413},"what-is-an-enterprise-agent-harness","enterprise agent harness"," is that application.",[626,1417,1419],{"id":1418},"do-we-wait-for-a-better-model-instead","Do we wait for a better model instead?",[189,1421,1422,1423,1426],{},"You still buy better models. You do not pause the ratchet. Stronger models attempt larger jobs and fail in new ways. Anthropic’s long-running work exists ",[473,1424,1425],{},"because"," models got good enough to outlast a window.",[626,1428,1430],{"id":1429},"how-do-we-start-this-quarter","How do we start this quarter?",[189,1432,1433,1434,1438],{},"Pick one job that already has a finish line. Encode guides. Attach one deterministic sensor. Add one hook that can refuse. Run it ten times. Every failure updates the harness. That is a ",[198,1435,1437],{"href":1436},"how-to-run-an-enterprise-ai-proof-of-value","proof of value"," for the practice, not a chat demo.",[626,1440,1442],{"id":1441},"how-does-nimbus-fit-without-becoming-the-definition","How does Nimbus fit without becoming the definition?",[189,1444,1445,1446,883,1448,883,1451,883,1454,1456,1457,1459],{},"Nimbus is an outer harness you can hire: ",[198,1447,382],{"href":32},[198,1449,1450],{"href":20},"teams",[198,1452,1453],{"href":40},"gates",[198,1455,1370],{"href":24},". Score it the way you score Claude Code: can you add a sensor, refuse a write, and replay who signed. ",[198,1458,1389],{"href":1388}," is the sheet.",[626,1461,1463],{"id":1462},"what-should-i-read-next","What should I read next?",[189,1465,1466,1470,1471,1475,1476,1480],{},[198,1467,1469],{"href":1468},"inner-vs-outer-agent-harness","Inner vs outer agent harness"," for the repo/company cut. ",[198,1472,1474],{"href":1473},"agent-harness-architecture","Agent harness architecture"," for the parts. ",[198,1477,1479],{"href":1478},"what-is-an-agent-harness","What is an agent harness"," if you still need the noun.",[204,1482,878],{"id":877},[189,1484,1485,383,1489,609],{},[198,1486,1488],{"href":1487},"what-is-an-enterprise-ai-operating-system","What is an enterprise AI operating system",[198,1490,1492],{"href":1491},"how-to-solve-ai-that-cannot-write-back-safely","How to solve AI that cannot write back safely",[204,1494,892],{"id":891},[277,1496,1497,1503,1510,1516,1522,1528,1534,1540,1547,1553,1559,1565,1571,1577,1583],{},[280,1498,1499],{},[198,1500,1502],{"href":961,"rel":1501},[227],"LangChain, Agents",[280,1504,1505],{},[198,1506,1509],{"href":1507,"rel":1508},"https://www.langchain.com/blog/the-anatomy-of-an-agent-harness",[227],"LangChain, The anatomy of an agent harness",[280,1511,1512],{},[198,1513,1515],{"href":977,"rel":1514},[227],"Böckeler, Harness engineering for coding agent users",[280,1517,1518],{},[198,1519,1521],{"href":1048,"rel":1520},[227],"Thoughtworks, Harness engineering and agent feedback",[280,1523,1524],{},[198,1525,1527],{"href":983,"rel":1526},[227],"Thoughtworks, Scaling the enterprise harness (podcast)",[280,1529,1530],{},[198,1531,1533],{"href":1084,"rel":1532},[227],"Thoughtworks, The operating system for enterprise AI",[280,1535,1536],{},[198,1537,1539],{"href":971,"rel":1538},[227],"Addy Osmani, Agent harness engineering",[280,1541,1542],{},[198,1543,1546],{"href":1544,"rel":1545},"https://www.oreilly.com/radar/agent-harness-engineering/",[227],"O’Reilly Radar, Agent harness engineering",[280,1548,1549],{},[198,1550,1552],{"href":995,"rel":1551},[227],"Anthropic, Building effective agents",[280,1554,1555],{},[198,1556,1558],{"href":1001,"rel":1557},[227],"Anthropic, Effective harnesses for long-running agents",[280,1560,1561],{},[198,1562,1564],{"href":1193,"rel":1563},[227],"Anthropic, Steering Claude Code",[280,1566,1567],{},[198,1568,1570],{"href":1060,"rel":1569},[227],"Claude Code, Hooks",[280,1572,1573],{},[198,1574,1576],{"href":225,"rel":1575},[227],"McKinsey, The state of AI in 2025",[280,1578,1579],{},[198,1580,1582],{"href":1111,"rel":1581},[227],"NIST AI RMF",[280,1584,1585],{},[198,1586,1152],{"href":1150,"rel":1587},[227],{"title":170,"searchDepth":171,"depth":171,"links":1589},[1590,1591,1592,1593,1594,1595,1596,1604,1605],{"id":482,"depth":171,"text":483},{"id":578,"depth":171,"text":579},{"id":1174,"depth":171,"text":1175},{"id":1265,"depth":171,"text":1266},{"id":1315,"depth":171,"text":1316},{"id":1374,"depth":171,"text":1375},{"id":767,"depth":171,"text":768,"children":1597},[1598,1599,1600,1601,1602,1603],{"id":1399,"depth":915,"text":1400},{"id":1406,"depth":915,"text":1407},{"id":1418,"depth":915,"text":1419},{"id":1429,"depth":915,"text":1430},{"id":1441,"depth":915,"text":1442},{"id":1462,"depth":915,"text":1463},{"id":877,"depth":171,"text":878},{"id":891,"depth":171,"text":892},"2026-08-24","Harness engineering is the 2026 practice of fixing the environment when an agent fails — tools, hooks, tests, and stops — instead of rewriting the prompt and hoping the next model call behaves.","/blog/what-is-harness-engineering",{"title":946,"description":1607},"blog/what-is-harness-engineering",[422,1612,1613,1614],"harness-engineering","agent-harness","evaluation","J7WO8MGgOG9nFjnGjs25nfHH5sz-UYeKtt3NvK0eueE",{"enabled":164,"message":1617,"linkLabel":93,"linkHref":94,"id":1618,"title":1619,"archived":164,"authors":165,"badge":165,"body":1620,"date":165,"definedTerm":165,"department":165,"description":170,"extension":173,"eyebrow":165,"faqHeader":165,"faqs":165,"footerBand":165,"headline":165,"image":165,"industry":165,"jobType":165,"listed":130,"location":165,"navigation":130,"openRoles":165,"pageLayout":165,"path":1624,"relatedHeading":165,"seo":1625,"series":165,"sitemap":164,"status":165,"stem":1626,"subhead":165,"tags":165,"video":165,"whyJoin":165,"workplaceType":165,"__hash__":1627},"We're hiring! Join the team building the Sentient Enterprise.","content/shared/hiring.md","Hiring banner",{"type":167,"value":1621,"toc":1622},[],{"title":170,"searchDepth":171,"depth":171,"links":1623},[],"/shared/hiring",{"title":1619,"description":170},"shared/hiring","1zs3boivKda1e-b-hAyuNcmZSKjZUAXmecnwHVgcHzk",{"fold":1629,"id":1633,"title":1634,"archived":164,"authors":165,"badge":165,"body":1635,"date":165,"definedTerm":165,"department":165,"description":170,"extension":173,"eyebrow":165,"faqHeader":165,"faqs":165,"footerBand":1639,"headline":165,"image":165,"industry":165,"jobType":165,"listed":130,"location":165,"navigation":130,"openRoles":165,"pageLayout":165,"path":1643,"relatedHeading":165,"seo":1644,"series":165,"sitemap":164,"status":165,"stem":1645,"subhead":165,"tags":165,"video":165,"whyJoin":165,"workplaceType":165,"__hash__":1646},{"headline":1630,"description":1631,"primaryLabel":8,"primaryTo":1632,"secondaryLabel":444,"secondaryTo":12},"Run frontier AI your business actually owns.","Governed agent swarms, 2,000+ integrations, and a knowledge graph that stays inside your walls. Free 7-day trial.","/checkout","content/shared/cta.md","Site CTAs",{"type":167,"value":1636,"toc":1637},[],{"title":170,"searchDepth":171,"depth":171,"links":1638},[],{"headline":1640,"description":1641,"primaryLabel":8,"primaryTo":1632,"secondaryLabel":1642,"secondaryTo":99},"See what governed AI looks like on your stack.","Connect your tools, run a workstream, and keep every decision on your ledger - free for 7 days.","Talk to our team","/shared/cta",{"title":1634,"description":170},"shared/cta","wz4AdRHnaYH021WMdWcHnZvHmkZJNKaNG4XGZfnFBtw",1788985847346]