[{"data":1,"prerenderedAt":2152},["ShallowReactive",2],{"site-nav-content":3,"blog:\u002Fblog\u002Fwhat-is-an-agent-harness":177,"blog-index-copy":882,"blog:\u002Fblog\u002Fwhat-is-an-agent-harness:surround":903,"hiring-banner-content":2121,"site-cta-content":2133},{"header":4,"productNav":9,"nav":42,"footer":61,"askAI":132,"id":161,"title":162,"archived":163,"authors":164,"badge":164,"body":165,"date":164,"definedTerm":164,"department":164,"description":169,"extension":172,"eyebrow":164,"faqHeader":164,"faqs":164,"footerBand":164,"headline":164,"image":164,"industry":164,"jobType":164,"listed":131,"location":164,"navigation":131,"openRoles":164,"pageLayout":164,"path":173,"relatedHeading":164,"seo":174,"series":164,"sitemap":163,"status":164,"stem":175,"subhead":164,"tags":164,"video":164,"whyJoin":164,"workplaceType":164,"__hash__":176},{"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","\u002Foverview","Seven layers. One closed loop.",{"label":15,"to":16,"description":17},"Conflux","\u002Fproduct\u002Fconflux","Where your team, workstreams, and agents meet.",{"label":19,"to":20,"description":21},"Agent Teams","\u002Fproduct\u002Fagent-teams","Specialist teams - governed from day one.",{"label":23,"to":24,"description":25},"Lifecycle Graph","\u002Fproduct\u002Flifecycle-graph","Intelligence that compounds across every interaction.",{"label":27,"to":28,"description":29},"Company Wiki","\u002Fproduct\u002Fwiki","Playbooks and policies where expertise stays.",{"label":31,"to":32,"description":33},"Workstreams","\u002Fproduct\u002Fworkstreams","From brief to signed-off deliverable on one canvas.",{"label":35,"to":36,"description":37},"Perception Console","\u002Fproduct\u002Fperception","Ask your whole business in plain English.",{"label":39,"to":40,"description":41},"Governance","\u002Fproduct\u002Fgovernance","Frontier AI you can actually sign off on.",[43,46,49,52,55,58],{"label":44,"to":45},"Models","\u002Fmodels",{"label":47,"to":48},"Pricing","\u002Fpricing",{"label":50,"to":51},"Integrations","\u002Fintegrations",{"label":53,"to":54},"Security","\u002Fsecurity",{"label":56,"to":57},"Partners","\u002Fpartners",{"label":59,"to":60},"Insights","\u002Fblog",{"productHeading":5,"companyHeading":62,"resourcesHeading":63,"legalHeading":64,"docsLabel":65,"docsUrl":66,"statementLines":67,"copyright":70,"companyLinks":71,"resourcesLinks":86,"legalLinks":102,"socialLinks":109,"bottomLinks":119},"Company","Resources","Legal","Docs","https:\u002F\u002Fdocs.gonimbus.ai",[68,69],"Stop training someone else's model.","Control your AI.","© 2026 Nimbus Intelligence, Inc. All rights reserved.",[72,73,74,75,76,78,81,84],{"label":47,"to":48},{"label":50,"to":51},{"label":53,"to":54},{"label":59,"to":60},{"label":77,"to":57},"Partner Program",{"label":79,"to":80},"Careers","\u002Fcareers",{"label":82,"to":83},"System status","\u002Fstatus",{"label":7,"to":85},"\u002Fcontact",[87,90,93,96,99],{"label":88,"to":89},"Glossary","\u002Fglossary",{"label":91,"to":92},"Compare","\u002Fcompare",{"label":94,"to":95},"Evaluate","\u002Fevaluate",{"label":97,"to":98},"Problems","\u002Fproblems",{"label":100,"to":101},"Use cases","\u002Fuse-cases",[103,106],{"label":104,"to":105},"Terms of Service","\u002Fterms",{"label":107,"to":108},"Privacy Policy","\u002Fprivacy",[110,113,116],{"label":111,"href":112},"LinkedIn","https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002Fgonimbusai\u002F",{"label":114,"href":115},"X","https:\u002F\u002Fx.com\u002Fgonimbusai",{"label":117,"href":118},"Instagram","https:\u002F\u002Fwww.instagram.com\u002Fgonimbus_ai\u002F",[120,122,124,127,128],{"label":121,"to":105},"Terms",{"label":123,"to":108},"Privacy",{"label":125,"to":126},"Compliance","\u002Fcompliance",{"label":82,"to":83},{"label":129,"to":130,"external":131},"LLMs.txt","\u002Fllms.txt",true,{"text":133,"prompt":134,"platforms":135},"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. Summarize the highlights from Nimbus's website: https:\u002F\u002Fgonimbus.ai",[136,141,146,151,156],{"name":137,"label":138,"icon":139,"hrefPrefix":140},"chatgpt","ChatGPT","simple-icons:openai","https:\u002F\u002Fchatgpt.com\u002F?prompt=",{"name":142,"label":143,"icon":144,"hrefPrefix":145},"perplexity","Perplexity","mdi:magnify","https:\u002F\u002Fwww.perplexity.ai\u002Fsearch\u002Fnew?q=",{"name":147,"label":148,"icon":149,"hrefPrefix":150},"grok","Grok","simple-icons:x","https:\u002F\u002Fx.com\u002Fi\u002Fgrok?text=",{"name":152,"label":153,"icon":154,"hrefPrefix":155},"claude","Claude","simple-icons:anthropic","https:\u002F\u002Fclaude.ai\u002Fnew?q=",{"name":157,"label":158,"icon":159,"hrefPrefix":160},"google-ai","Google AI","simple-icons:google","https:\u002F\u002Fwww.google.com\u002Fsearch?udm=50&aep=11&q=","content\u002Fshared\u002Fnav.md","Site navigation",false,null,{"type":166,"value":167,"toc":168},"minimark",[],{"title":169,"searchDepth":170,"depth":170,"links":171},"",2,[],"md","\u002Fshared\u002Fnav",{"title":162,"description":169},"shared\u002Fnav","Q7sDe7TuGEwhwUMamyeOxjvGlsUaF3Iay9VTW0KaE_U",{"id":178,"title":179,"archived":163,"authors":180,"badge":184,"body":186,"date":858,"definedTerm":164,"department":164,"description":859,"extension":172,"eyebrow":164,"faqHeader":860,"faqs":863,"footerBand":164,"headline":164,"image":164,"industry":164,"jobType":164,"listed":131,"location":164,"navigation":131,"openRoles":164,"pageLayout":164,"path":873,"relatedHeading":164,"seo":874,"series":875,"sitemap":131,"status":164,"stem":876,"subhead":164,"tags":877,"video":164,"whyJoin":164,"workplaceType":164,"__hash__":881},"content\u002Fblog\u002Fwhat-is-an-agent-harness.md","What is an Agent Harness",[181],{"name":182,"to":183},"Nimbus Research","https:\u002F\u002Fgonimbus.ai",{"label":185},"Explainer",{"type":166,"value":187,"toc":839},[188,207,215,227,245,250,359,381,385,392,395,412,425,442,449,453,456,471,482,500,513,523,534,550,554,557,560,568,575,594,614,618,630,639,650,657,661,666,669,673,676,680,688,692,695,699,705,709,716,720,733,737,749,753],[189,190,191,192,196,197,202,203,206],"p",{},"An ",[193,194,195],"strong",{},"agent harness"," is the software around a large language model that turns next-token prediction into work: tools, memory, a loop, permissions, and a stop. ",[198,199,201],"a",{"href":200},"https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Flangchain\u002Fagents","LangChain’s 2026 documentation"," writes the equation in plain type: ",[193,204,205],{},"Agent = Model + Harness",". The model reasons. The harness is everything else.",[189,208,209,210,214],{},"That sentence is not marketing. An unaided model is stateless. It produces text. It cannot keep a file, call Salesforce, fail a linter, or refuse a write. The ",[198,211,213],{"href":212},"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FAgent_harness","Wikipedia entry on agent harnesses"," records the same split, and notes that the UK’s AI Security Institute already described an AI agenthe model plus scaffolding in 2023. The industry spent two years arguing about which model was smartest. In 2026 it started arguing about which environment the model was sitting in.",[189,216,217,221,222,226],{},[198,218,220],{"href":219},"https:\u002F\u002Fwww.databricks.com\u002Fblog\u002Fai-harness","Databricks"," uses a body-and-brain analogy: the model is the brain; the harness is the body and the workspace. ",[198,223,225],{"href":224},"https:\u002F\u002Fwww.langchain.com\u002Fblog\u002Fthe-anatomy-of-an-agent-harness","LangChain’s anatomy post"," is more mechanical. A harness is every piece of code, configuration, and execution logic that is not the model itself. A raw model is not an agent. It becomes one when a harness gives it state, tool execution, feedback loops, and constraints that do not depend on the model’s mood.",[189,228,229,230,234,235,239,240,244],{},"This article is the definition. ",[198,231,233],{"href":232},"\u002Fblog\u002Fwhat-is-harness-engineering\u002F","What is harness engineering"," is the practice of tightening that environment when the agent fails. ",[198,236,238],{"href":237},"\u002Fblog\u002Finner-vs-outer-agent-harness\u002F","Inner vs outer agent harness"," is the cut between a repo and a company. An ",[198,241,243],{"href":242},"\u002Fblog\u002Fwhat-is-an-enterprise-agent-harness\u002F","enterprise agent harness"," is the outer case: operators, signers, a ledger.",[246,247,249],"h2",{"id":248},"words-youll-hear","Words you’ll hear",[251,252,253,260,266,272,283,299,319,334,343],"ul",{},[254,255,256,259],"li",{},[193,257,258],{},"Harness \u002F scaffolding."," Same object, two eras. Scaffolding is the 2023–2024 research word. Harness is the 2026 product word. Both mean the runtime around the weights.",[254,261,262,265],{},[193,263,264],{},"Agent."," The composed system. Not the model. Not the chat UI. If you can swap the model and the job still runs, you were looking at the harness.",[254,267,268,271],{},[193,269,270],{},"Loop."," Plan, act, observe, repeat — the ReAct-shaped cycle popularised in 2022 and now owned by the harness, not by the prompt. The harness dispatches the tool, returns the result, and decides whether to continue.",[254,273,274,277,278,282],{},[193,275,276],{},"Stop."," Budget, max steps, empty retrieval, tool error, human cancel, wait-for-named-signer. “The model says it is done” is a suggestion. ",[198,279,281],{"href":280},"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fbuilding-effective-agents","Anthropic’s note on building effective agents"," is honest about this: encoding the job and deciding what “done” means is the boring part that actually matters.",[254,284,285,288,289,293,294,298],{},[193,286,287],{},"Tools \u002F skills \u002F MCP."," Hands. The ",[198,290,292],{"href":291},"https:\u002F\u002Fmodelcontextprotocol.io\u002Fspecification\u002F2025-11-25\u002Findex","Model Context Protocol"," is a common plug so hosts can call the same servers. Plumbing. A plug is not a permission model. See ",[198,295,297],{"href":296},"\u002Fblog\u002Fwhat-is-model-context-protocol\u002F","What is Model Context Protocol",".",[254,300,301,304,305,309,310,314,315,318],{},[193,302,303],{},"Hooks \u002F middleware."," Deterministic intercepts on the loop. ",[198,306,308],{"href":307},"https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fhooks","Claude Code hooks"," run shell or HTTP at ",[311,312,313],"code",{},"PreToolUse"," and can block with exit code 2. LangChain middleware is the same instinct in a library. A line in ",[311,316,317],{},"CLAUDE.md"," is advice. A hook is a gate.",[254,320,321,324,325,329,330,333],{},[193,322,323],{},"Guides and sensors."," ",[198,326,328],{"href":327},"https:\u002F\u002Fmartinfowler.com\u002Farticles\u002Fharness-engineering.html","Birgitta Böckeler’s framing on martinfowler.com",": feed-forward context (conventions, architecture, ",[311,331,332],{},"AGENTS.md",") versus feedback (linters, tests, reviewers). A harness that only prompts is half a harness.",[254,335,336,339,340,298],{},[193,337,338],{},"Inner harness."," Coding agents: Claude Code, Cursor, Codex. Workspace is a repository. Tests are the eval. See ",[198,341,342],{"href":237},"inner vs outer",[254,344,345,348,349,353,354,358],{},[193,346,347],{},"Outer \u002F enterprise harness."," Operators. Connectors, ",[198,350,352],{"href":351},"\u002Fblog\u002Fwhat-is-an-ai-workstream\u002F","workstreams",", ",[198,355,357],{"href":356},"\u002Fblog\u002Fwhat-is-write-back-governance\u002F","write-back",", a ledger. Workspace is the company. A passing unit test does not prove a CRM write was authorised.",[189,360,361,362,353,366,353,370,353,373,353,377,380],{},"Nimbus is one outer harness: ",[198,363,365],{"href":364},"\u002Fproduct\u002Fwiki\u002F","wiki",[198,367,369],{"href":368},"\u002Fproduct\u002Fagent-teams\u002F","agent teams",[198,371,352],{"href":372},"\u002Fproduct\u002Fworkstreams\u002F",[198,374,376],{"href":375},"\u002Fproduct\u002Fgovernance\u002F","governance",[198,378,23],{"href":379},"\u002Fproduct\u002Flifecycle-graph\u002F",". Claude Code is a strong inner harness. Calling either “an agent” without naming the harness is how RFPs buy a model and inherit someone else’s loop.",[246,382,384],{"id":383},"why-you-should-care","Why you should care",[189,386,387,391],{},[198,388,390],{"href":389},"https:\u002F\u002Fwww.mckinsey.com\u002Fcapabilities\u002Fquantumblack\u002Four-insights\u002Fthe-state-of-ai","McKinsey’s 2025 State of AI survey"," is the scale gap in one chart: most organisations use AI in at least one function; far fewer have begun to scale. Copilots produce usage. Harnesses produce jobs that finish under a stop. If your programme is “we rolled out ChatGPT Enterprise,” you have licensed a model surface. You have not yet chosen a harness for the work that writes back.",[189,393,394],{},"It affects you if:",[251,396,397,400,403,406,409],{},[254,398,399],{},"the job is multi-step and tool-using, not a single completion",[254,401,402],{},"a live system can change (CRM, ERP, repo, ticket queue)",[254,404,405],{},"someone will ask, six months later, why a field or a file changed",[254,407,408],{},"you need to swap models without rewriting every tool",[254,410,411],{},"you already noticed that a better model still skips the linter, invents a policy, or pastes into Salesforce",[189,413,414,415,419,420,424],{},"In February 2024 a British Columbia tribunal held Air Canada responsible for a chatbot that invented a bereavement-fare policy. ",[198,416,418],{"href":417},"https:\u002F\u002Fwww.cbc.ca\u002Fnews\u002Fcanada\u002Fbritish-columbia\u002Fair-canada-chatbot-lawsuit-1.7116416","CBC reported"," that the airline’s argument — the chatbot is a separate legal entity — failed. That failure is a missing harness, not a missing model: no quote, no signer, no stop before a commitment left the building. In 2023 a New York court ",[198,421,423],{"href":422},"https:\u002F\u002Fwww.reuters.com\u002Flegal\u002Fnew-york-lawyers-sanctioned-using-fake-chatgpt-cases-legal-brief-2023-06-22\u002F","sanctioned lawyers"," who filed ChatGPT-invented cases. Ungated generation reached a system of record. CRM writes are the operational twin with money attached.",[189,426,427,431,432,436,437,441],{},[198,428,430],{"href":429},"https:\u002F\u002Fwww.nist.gov\u002Fitl\u002Fai-risk-management-framework","NIST’s AI RMF"," organises Govern, Map, Measure, Manage. ",[198,433,435],{"href":434},"https:\u002F\u002Fwww.iso.org\u002Fstandard\u002F42001","ISO\u002FIEC 42001"," is an AI ",[438,439,440],"em",{},"management system"," standard. Neither is implemented by a system prompt that says “be careful.” They are implemented by a runtime that can refuse a tool call.",[189,443,444,448],{},[198,445,447],{"href":446},"https:\u002F\u002Faddyosmani.com\u002Fblog\u002Fagent-harness-engineering\u002F","Addy Osmani’s 2026 write-up"," states the engineering claim operators keep rediscovering: a decent model with a great harness beats a great model with a bad harness. When the agent does something dumb, the default instinct is to blame the weights. Harness engineering treats most of those failures as configuration. That is the rest of this cluster.",[246,450,452],{"id":451},"what-a-harness-actually-contains","What a harness actually contains",[189,454,455],{},"LangChain’s anatomy and Databricks’s list converge on the same parts. You can inspect each one before you buy a product or assemble a library.",[189,457,458,461,462,466,467,470],{},[193,459,460],{},"The loop."," The harness owns plan → act → observe. It executes the tool. It feeds the result back. It enforces max steps and a cost budget so a stuck agent cannot run forever. ",[198,463,465],{"href":464},"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-harnesses-for-long-running-agents","Anthropic’s long-running harness note"," shows why this is not a prompt: tasks that outlast one context window need an initializer, incremental sessions, git commits, and a progress file the ",[438,468,469],{},"next"," session can read. The model does not remember. The environment does.",[189,472,473,476,477,481],{},[193,474,475],{},"Tools and execution."," Search, shell, apply_patch, browser, CRM, ERP. The model proposes a call. The harness runs it in a sandbox or against an adapter, handles timeouts, and returns structured results. A generic HTTP tool with a production token is not a harness. It is a confused deputy. ",[198,478,480],{"href":479},"https:\u002F\u002Fgenai.owasp.org\u002Fllm-top-10\u002F","OWASP’s Top 10 for LLM applications"," still applies: excessive agency and unbounded tool use are design failures, not model quirks.",[189,483,484,487,488,490,491,495,496,298],{},[193,485,486],{},"Context and memory."," Working memory is the current window. Session state is progress for this job. Durable memory is files, ",[311,489,332],{},", a wiki, or a graph — something that survives compaction. Anthropic’s initializer\u002Fcoding-agent split is a memory design: feature lists and commits as cross-session state. A company that stores “what we approved” only in Slack search does not have durable memory for operations. See ",[198,492,494],{"href":493},"\u002Fblog\u002Fwhat-is-a-lifecycle-graph\u002F","What is a lifecycle graph"," and ",[198,497,499],{"href":498},"\u002Fblog\u002Fwhat-is-a-company-wiki-for-ai-agents\u002F","What is a company wiki for AI agents",[189,501,502,505,506,508,509,512],{},[193,503,504],{},"Permissions and hooks."," Who may call which tool, with which identity, on which object. Claude Code’s ",[311,507,313],{}," hook can deny Bash regardless of what the model intended. That is the inner version of ",[198,510,511],{"href":356},"write-back governance",": the write API is unreachable until a named role signs a quoted payload. A prompt that says “ask Legal first” is not this layer. The model can forget. The user can paste anyway.",[189,514,515,518,519,298],{},[193,516,517],{},"Feedback."," Compilers, tests, linters, schema validators, human review. Böckeler’s sensors. Without them the loop is open: the model reports success and the harness believes it. Terminal-Bench and SWE-bench exist because coding harnesses can grade against an environment. Enterprise writes need an equivalent: did the signed payload match what executed. See ",[198,520,522],{"href":521},"\u002Fblog\u002Feval-loops-for-enterprise-agent-harnesses\u002F","eval loops for enterprise agent harnesses",[189,524,525,528,529,533],{},[193,526,527],{},"Orchestration."," Subagents, hand-offs, model routing. Optional until the job already splits in the organisation. ",[198,530,532],{"href":531},"\u002Fblog\u002Fwhat-is-multi-agent-ai\u002F","Multi-agent AI"," is the pattern and the hiring object. A harness that spawns specialists without a stop is a faster way to share a production login.",[189,535,191,536,540,541,544,545,549],{},[198,537,539],{"href":538},"\u002Fblog\u002Fwhat-is-an-agentic-workflow\u002F","agentic workflow"," is a designed sequence with business stops. The harness is the runtime that can actually run that sequence. Mixing those two words is how demos skip isolation. A ",[198,542,543],{"href":351},"workstream"," is the company object that hosts the job: brief, connectors, people, budget. In Nimbus the workstream is that folder; the harness is wiki + teams + connectors + gates + graph sitting around whichever model ",[198,546,548],{"href":547},"\u002Fmodels\u002F","routing"," picks for the step.",[246,551,553],{"id":552},"what-is-not-a-harness","What is not a harness",[189,555,556],{},"A chat window with plugins. The human is still the message bus, the permission system, and the audit log.",[189,558,559],{},"A system prompt. Advice inside the window. Useful. Not a stop.",[189,561,562,563,567],{},"A policy PDF. ",[198,564,566],{"href":565},"\u002Fblog\u002Fwhat-is-ai-governance\u002F","What is AI governance"," is a management claim. A harness is whether an unapproved write is impossible.",[189,569,570,571,574],{},"MCP on its own. A standard plug. See ",[198,572,573],{"href":296},"what is Model Context Protocol",". If the server can PATCH Salesforce from natural language, you built a bypass.",[189,576,577,578,585,586,589,590,298],{},"A framework on its own. ",[198,579,581,582],{"href":580},"https:\u002F\u002Fwww.langchain.com\u002Fblog\u002Fhow-to-build-a-custom-agent-harness","LangChain’s ",[311,583,584],{},"create_agent"," is a way to ",[438,587,588],{},"assemble"," a harness. CrewAI, LangGraph, and Pydantic AI are in the same neighbourhood. You still have to choose tools, stops, and identity. See ",[198,591,593],{"href":592},"\u002Fblog\u002Fagent-harness-vs-agent-framework\u002F","agent harness vs agent framework",[189,595,596,597,495,601,605,606,495,610,298],{},"A copilot seat. ",[198,598,600],{"href":599},"https:\u002F\u002Fopenai.com\u002Fbusiness\u002Fchatgpt-enterprise\u002F","ChatGPT Enterprise",[198,602,604],{"href":603},"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fmicrosoft-365\u002Fcopilot","Microsoft 365 Copilot"," are excellent personal surfaces. They are not, by default, a company loop with fail-closed writes. See ",[198,607,609],{"href":608},"\u002Fblog\u002Fcollaborative-ai-and-personal-assistants\u002F","collaborative AI and personal assistants",[198,611,613],{"href":612},"\u002Fblog\u002Fhow-to-choose-between-a-coding-harness-and-an-enterprise-harness\u002F","how to choose between a coding harness and an enterprise harness",[246,615,617],{"id":616},"how-this-shows-up-in-products","How this shows up in products",[189,619,620,623,624,626,627,629],{},[193,621,622],{},"Coding harnesses."," Claude Code, Cursor, Codex, open shells like OpenHands. Workspace is a checkout. ",[311,625,317],{}," \u002F ",[311,628,332],{}," are guides. Hooks, tests, and CI are sensors. Eval is SWE-bench or Terminal-Bench, or your own suite. These are the right shape for software.",[189,631,632,635,636,638],{},[193,633,634],{},"Library harnesses."," LangChain ",[311,637,584],{},", Deep Agents, LangGraph graphs. You compose the loop in code. You own production identity. Good when the job is yours to engineer. A liability when operators are expected to “just add Salesforce.”",[189,640,641,644,645,649],{},[193,642,643],{},"Enterprise \u002F outer harnesses."," Palantir AIP, Salesforce Agentforce, and self-service OS-class products such as Nimbus. The workspace is a job with connectors and people, not a git root. The interesting stop is a named signer on a quoted write, not a green test. ",[198,646,648],{"href":647},"\u002Fblog\u002Fhow-to-evaluate-an-agent-harness\u002F","How to evaluate an agent harness"," is the buying sheet.",[189,651,652,653,656],{},"Nimbus’s mapping is deliberate and not unique as a ",[438,654,655],{},"category",": Perception orients, Conflux collaborates, agent teams run, governance quotes, the graph records. You can score that mapping against the parts above. You should score AIP and Agentforce the same way. Category names do not substitute for a failed write.",[246,658,660],{"id":659},"questions-people-actually-ask","Questions people actually ask",[662,663,665],"h3",{"id":664},"is-the-model-the-agent","Is the model the agent?",[189,667,668],{},"No. The agent is model plus harness. Shopping for a model is shopping for a chip. Shopping for a harness is shopping for how work finishes.",[662,670,672],{"id":671},"do-i-need-a-harness-for-a-single-prompt","Do I need a harness for a single prompt?",[189,674,675],{},"No. A completion does not need a loop. Multi-step tool use does. Long-running work that outlasts one window does. Writes to live systems do.",[662,677,679],{"id":678},"is-rag-a-harness","Is RAG a harness?",[189,681,682,683,687],{},"Retrieval is a tool and a memory pattern inside a step. ",[198,684,686],{"href":685},"\u002Fblog\u002Fwhat-is-enterprise-rag\u002F","Enterprise RAG"," does not dispatch tools, enforce a signer, or persist a decision. Useful. Incomplete.",[662,689,691],{"id":690},"can-i-just-use-mcp-as-my-harness","Can I just use MCP as my harness?",[189,693,694],{},"You can use MCP as the plug. You still need identity, scope, quoting, and a stop. The spec does not require those.",[662,696,698],{"id":697},"will-a-better-model-shrink-the-harness","Will a better model shrink the harness?",[189,700,701,704],{},[198,702,703],{"href":446},"Osmani"," and Anthropic’s long-running work both say the ceiling moves. Tasks that were unreachable come into play and bring new failure modes. Stronger models still do not know your signer, your budget, or your CRM field map.",[662,706,708],{"id":707},"how-is-this-different-from-an-enterprise-ai-os","How is this different from an enterprise AI OS?",[189,710,191,711,715],{},[198,712,714],{"href":713},"\u002Fblog\u002Fwhat-is-an-enterprise-ai-operating-system\u002F","enterprise AI operating system"," is the company-shaped product: wiki, workstreams, teams, gates, ledger. A harness is the runtime idea underneath — including coding harnesses that are not an OS. Nimbus is an OS-class outer harness. Claude Code is not an OS. Both are harnesses.",[662,717,719],{"id":718},"what-should-i-read-next","What should I read next?",[189,721,722,724,725,729,730,732],{},[198,723,233],{"href":232}," for the practice. ",[198,726,728],{"href":727},"\u002Fblog\u002Fagent-harness-architecture\u002F","Agent harness architecture"," for the parts in one diagram. ",[198,731,648],{"href":647}," before a vendor demo.",[246,734,736],{"id":735},"related-reading","Related reading",[189,738,739,353,742,745,746,298],{},[198,740,741],{"href":538},"What is an agentic workflow",[198,743,744],{"href":531},"What is multi-agent AI",", and ",[198,747,748],{"href":356},"What is write-back governance",[246,750,752],{"id":751},"sources","Sources",[251,754,755,760,765,770,775,780,785,790,795,800,805,810,814,819,824,829,834],{},[254,756,757],{},[198,758,759],{"href":200},"LangChain, Agents (Agent = Model + Harness)",[254,761,762],{},[198,763,764],{"href":224},"LangChain, The anatomy of an agent harness",[254,766,767],{},[198,768,769],{"href":580},"LangChain, How to build a custom agent harness",[254,771,772],{},[198,773,774],{"href":212},"Wikipedia, Agent harness",[254,776,777],{},[198,778,779],{"href":219},"Databricks, What is an AI agent harness?",[254,781,782],{},[198,783,784],{"href":280},"Anthropic, Building effective agents",[254,786,787],{},[198,788,789],{"href":464},"Anthropic, Effective harnesses for long-running agents",[254,791,792],{},[198,793,794],{"href":327},"Böckeler, Harness engineering for coding agent users",[254,796,797],{},[198,798,799],{"href":446},"Addy Osmani, Agent harness engineering",[254,801,802],{},[198,803,804],{"href":389},"McKinsey, The state of AI in 2025",[254,806,807],{},[198,808,809],{"href":429},"NIST AI Risk Management Framework",[254,811,812],{},[198,813,435],{"href":434},[254,815,816],{},[198,817,818],{"href":417},"CBC, Air Canada chatbot lawsuit",[254,820,821],{},[198,822,823],{"href":422},"Reuters, New York lawyers sanctioned over ChatGPT citations",[254,825,826],{},[198,827,828],{"href":479},"OWASP Top 10 for LLM applications",[254,830,831],{},[198,832,833],{"href":291},"Model Context Protocol specification (2025-11-25)",[254,835,836],{},[198,837,838],{"href":307},"Claude Code, Hooks",{"title":169,"searchDepth":170,"depth":170,"links":840},[841,842,843,844,845,846,856,857],{"id":248,"depth":170,"text":249},{"id":383,"depth":170,"text":384},{"id":451,"depth":170,"text":452},{"id":552,"depth":170,"text":553},{"id":616,"depth":170,"text":617},{"id":659,"depth":170,"text":660,"children":847},[848,850,851,852,853,854,855],{"id":664,"depth":849,"text":665},3,{"id":671,"depth":849,"text":672},{"id":678,"depth":849,"text":679},{"id":690,"depth":849,"text":691},{"id":697,"depth":849,"text":698},{"id":707,"depth":849,"text":708},{"id":718,"depth":849,"text":719},{"id":735,"depth":170,"text":736},{"id":751,"depth":170,"text":752},"2026-08-24","An agent harness is everything around a model that lets it do work: tools, memory, permissions, loops, and stops.",{"eyebrow":861,"title":862},"Short answers","Everything around the model",[864,867,870],{"question":865,"answer":866},"What is an agent harness?","Everything around a model that lets it do work: tools, memory, permissions, loops, and stops. Agent equals model plus harness.",{"question":868,"answer":869},"Is a chat window with plugins a harness?","No. Plugins without a stop, an identity, and a record are still a chat.",{"question":871,"answer":872},"What does the model do, and what does the harness do?","The model proposes. The harness decides which tools exist, what may be written, and when the loop ends.","\u002Fblog\u002Fwhat-is-an-agent-harness",{"title":179,"description":859},"explainer","blog\u002Fwhat-is-an-agent-harness",[875,878,879,880],"agent-harness","agents","harness-engineering","nEbkYgxLqx9029Xl-euoJSJGlRL_AZRKSh0pOAnQr80",{"hero":883,"id":885,"title":886,"archived":163,"authors":164,"badge":164,"body":887,"date":164,"definedTerm":164,"department":164,"description":891,"extension":172,"eyebrow":892,"faqHeader":164,"faqs":164,"footerBand":893,"headline":164,"image":164,"industry":164,"jobType":164,"listed":131,"location":164,"navigation":131,"openRoles":164,"pageLayout":164,"path":60,"relatedHeading":899,"seo":900,"series":164,"sitemap":131,"status":164,"stem":901,"subhead":164,"tags":164,"video":164,"whyJoin":164,"workplaceType":164,"__hash__":902},{"filename":884},"u2221455217_Flat_design_of_a_futuristic_minimalist_landscape__5d589295-cdea-4ea9-a262-be766881accf_1.png","content\u002Fblog\u002Findex.md","Exploring the future of intelligence.",{"type":166,"value":888,"toc":889},[],{"title":169,"searchDepth":170,"depth":170,"links":890},[],"Deep dives into pre-cognitive intelligence, sentient enterprises, and the evolving landscape of AI-driven business transformation.","Latest Research",{"headline":894,"description":895,"primaryLabel":896,"primaryTo":897,"secondaryLabel":898,"secondaryTo":12},"Stay at the frontier.","Subscribe for product updates and new insights.","Subscribe","\u002Fnewsletter","Explore the platform","More research",{"title":886,"description":891},"blog\u002Findex","BFSWGYO9bcTlaulivKYWyg08_DJHsdGg3OC6g_CG1Hw",[904,1511],{"id":905,"title":906,"archived":163,"authors":907,"badge":909,"body":910,"date":858,"definedTerm":164,"department":164,"description":1492,"extension":172,"eyebrow":164,"faqHeader":1493,"faqs":1495,"footerBand":164,"headline":164,"image":164,"industry":164,"jobType":164,"listed":131,"location":164,"navigation":131,"openRoles":164,"pageLayout":164,"path":1505,"relatedHeading":164,"seo":1506,"series":875,"sitemap":131,"status":164,"stem":1507,"subhead":164,"tags":1508,"video":164,"whyJoin":164,"workplaceType":164,"__hash__":1510},"content\u002Fblog\u002Fwhat-is-an-enterprise-agent-harness.md","What is an Enterprise Agent Harness",[908],{"name":182,"to":183},{"label":185},{"type":166,"value":911,"toc":1473},[912,925,936,948,950,1014,1031,1033,1052,1054,1071,1093,1106,1110,1118,1134,1150,1159,1173,1181,1190,1201,1207,1211,1216,1222,1232,1235,1245,1249,1260,1267,1270,1273,1281,1285,1288,1294,1303,1306,1310,1326,1332,1334,1338,1341,1345,1348,1352,1355,1359,1366,1370,1377,1379,1388,1390,1397,1399],[189,913,191,914,916,917,920,921,924],{},[193,915,243],{}," is the outer runtime that lets a model work on company jobs: policy it actually loads, connectors with least privilege, a loop that can stop for a named signer, and a record you can query after the people change. Buyers also ask for a ",[193,918,919],{},"business AI harness"," or a ",[193,922,923],{},"general enterprise harness",". Those phrases name the same object: a runtime for operators and systems of record, not a coding harness and not a chat with every production login.",[189,926,927,928,930,931,935],{},"It is still ",[198,929,205],{"href":200},". The workspace is not a git root. The sensor is not only pytest. The stop is not only max steps. ",[198,932,934],{"href":933},"https:\u002F\u002Fwww.thoughtworks.com\u002Finsights\u002Farticles\u002Foperating-system-enterprise-ai","Thoughtworks"," calls the missing piece an organisational harness: identity, ownership, economics, and learning around whatever builder harnesses (Claude Code, Cursor, LangChain graphs) teams already bought. An enterprise agent harness is that layer made operable — whether you assemble it or hire it.",[189,937,938,941,942,944,945,947],{},[198,939,940],{"href":237},"Inner vs outer"," is the cut. This page is the outer object in full. An ",[198,943,714],{"href":713}," is the product category that usually ships it: wiki, ",[198,946,352],{"href":351},", teams, gates, ledger. You can have OS-class products (Nimbus, Palantir AIP, Salesforce Agentforce) and still fail the harness test if writes are a boolean on an API key. You can assemble an enterprise harness in LangGraph and pass the test. The noun is the runtime properties, not the logo.",[246,949,249],{"id":248},[251,951,952,960,966,972,981,987,995,1004],{},[254,953,954,957,958,298],{},[193,955,956],{},"Outer harness."," Company workspace. See ",[198,959,342],{"href":237},[254,961,962,965],{},[193,963,964],{},"Organizational harness."," Thoughtworks’ fourth layer after model, builder harness, and user harness. Governance architecture, not another markdown file.",[254,967,968,971],{},[193,969,970],{},"Workstream."," Isolation domain: roster, connectors, budget, finish line. The job folder. Not a chat title.",[254,973,974,977,978,298],{},[193,975,976],{},"Write quoting."," The human sees the change in the language of the live system before sign-off. ",[198,979,980],{"href":356},"Write-back governance",[254,982,983,986],{},[193,984,985],{},"Fail-closed."," Missing approval, detached grant, or down interceptor means nothing mutates. Fail-open is a faster incident.",[254,988,989,992,993,298],{},[193,990,991],{},"Ledger \u002F Lifecycle Graph."," AI operations events: brief, agents, policy version, signer, payload. Distinct from the warehouse’s business events. See ",[198,994,494],{"href":493},[254,996,997,1000,1001,298],{},[193,998,999],{},"SWE-bench \u002F Terminal-Bench."," Inner evals. Useful for engineering vendors. Not a SOX control. ",[198,1002,1003],{"href":521},"Eval loops",[254,1005,1006,1009,1010,298],{},[193,1007,1008],{},"Forward-deployed programme."," Vendor engineers for months. AIP at scale. Capability can be real. Time-to-value is staffing. ",[198,1011,1013],{"href":1012},"\u002Fevaluate\u002F","Self-service vs forward-deployed",[189,1015,1016,1017,353,1019,353,1021,353,1023,353,1025,353,1028,1030],{},"Nimbus is one self-service enterprise harness: ",[198,1018,365],{"href":364},[198,1020,352],{"href":372},[198,1022,369],{"href":368},[198,1024,376],{"href":375},[198,1026,1027],{"href":379},"graph",[198,1029,548],{"href":547},". Score it as an example of the shape, next to AIP and Agentforce, not as the definition of the category.",[246,1032,384],{"id":383},[189,1034,1035,1038,1039,1042,1043,1046,1047,1051],{},[198,1036,1037],{"href":389},"McKinsey’s 2025 State of AI"," keeps separating ",[438,1040,1041],{},"use"," from ",[438,1044,1045],{},"scale",". Copilots and coding harnesses can produce the first. Enterprise harnesses are how writes to systems of record become the second without becoming ",[198,1048,1050],{"href":1049},"\u002Fblog\u002Fwhat-is-shadow-ai\u002F","shadow AI"," in the CRM.",[189,1053,394],{},[251,1055,1056,1059,1062,1065,1068],{},[254,1057,1058],{},"RevOps, Legal, and Finance must share a job, not a Slack channel of screenshots",[254,1060,1061],{},"Salesforce or NetSuite can change because a model proposed it",[254,1063,1064],{},"last quarter’s pricing chat is unrecoverable",[254,1066,1067],{},"security cannot list the AI actors that may write",[254,1069,1070],{},"the vendor demo is a SWE-bench plot and a “we have MCP”",[189,1072,1073,1074,1077,1078,1082,1083,1087,1088,1092],{},"In 2024 Air Canada was held to a chatbot’s invented policy (",[198,1075,1076],{"href":417},"CBC","). That is an outer-harness failure: a commitment left the building without a quote or a signer. ",[198,1079,1081],{"href":1080},"https:\u002F\u002Feur-lex.europa.eu\u002Feli\u002Freg\u002F2016\u002F679\u002Foj","GDPR"," constrains personal data in payloads. ",[198,1084,1086],{"href":1085},"https:\u002F\u002Fwww.sec.gov\u002Fabout\u002Flaws.shtml","Sarbanes–Oxley"," constrains who may change revenue truth. ",[198,1089,1091],{"href":1090},"https:\u002F\u002Feur-lex.europa.eu\u002Feli\u002Freg\u002F2024\u002F1689\u002Foj","EU AI Act"," Article 14 wants people who can interpret, interrupt, and leave a record. A coding-agent hook that formats Python does not satisfy those.",[189,1094,1095,495,1098,1100,1101,1105],{},[198,1096,1097],{"href":429},"NIST AI RMF",[198,1099,435],{"href":434}," assume operational controls, not a slide titled governance. ",[198,1102,1104],{"href":1103},"https:\u002F\u002Foecd.ai\u002Fen\u002Fai-principles","OECD AI Principles"," are a board checklist. They do not implement a gate. The harness does.",[246,1107,1109],{"id":1108},"what-enterprise-adds-to-a-harness","What “enterprise” adds to a harness",[189,1111,1112,1113,1117],{},"Start from ",[198,1114,1116],{"href":1115},"\u002Fblog\u002Fwhat-is-an-agent-harness\u002F","what a harness contains"," — loop, tools, memory, permissions, feedback, orchestration — and raise the bar.",[189,1119,1120,1123,1124,1126,1127,1130,1131,1133],{},[193,1121,1122],{},"Policy that loads."," Inner harnesses inject ",[311,1125,332],{},". Enterprise harnesses inject asserted company policy for ",[438,1128,1129],{},"this"," job, versioned. A Drive dump is not policy. A ",[198,1132,365],{"href":498}," that agents cite, with the revision on the run, is. If Legal’s discount cap lives only in a PDF nobody attached, the model will invent a number. That is not hallucination as a personality. That is a missing guide.",[189,1135,1136,1139,1140,1144,1145,1147,1148,298],{},[193,1137,1138],{},"Connectors as grants, not a toolbox."," Default read. Write is a separate plane. Least privilege is a workstream property. ",[198,1141,1143],{"href":1142},"\u002Fintegrations\u002F","Connector architecture",". MCP may be the plug; it must inherit the grant. ",[198,1146,297],{"href":296},". A Finance team assigned to a GTM-only stream still must not reach ERP “because it is Finance.” ",[198,1149,744],{"href":531},[189,1151,1152,1155,1156,1158],{},[193,1153,1154],{},"A hiring object for operators."," Not a folder of personal GPTs. A mandate, required systems, approval triggers — ",[198,1157,369],{"href":531}," as a roster. Nimbus ships functional teams on that roster; AIP and Agentforce have their own packaging. The test is: can an operator inspect the mandate and the required systems before assign.",[189,1160,1161,324,1164,1168,1169,1172],{},[193,1162,1163],{},"Human wait as a step.",[198,1165,1167],{"href":1166},"\u002Fblog\u002Fwhat-is-human-in-the-loop-ai\u002F","HITL"," is not a kill switch in a dashboard. It is quoted payload, named role, fail-closed adapter. ",[198,1170,1171],{"href":1166},"HITL approval architecture",". Soft \u002F Hard \u002F Critical matched to blast radius. A six-month zero-reject rate on CRM writes is a finding.",[189,1174,1175,1178,1179,298],{},[193,1176,1177],{},"A ledger of AI operations."," Who briefed, which team, which wiki revision, who signed, what executed. Exportable without the vendor in the room. The warehouse is not this ledger. ",[198,1180,494],{"href":493},[189,1182,1183,1186,1187,298],{},[193,1184,1185],{},"Evals that match the job."," Did the executed write match the signed quote. Can you replay. Inner leaderboards are a vendor quality signal for coding. They are not the enterprise eval. See ",[198,1188,1189],{"href":521},"eval loops",[189,1191,1192,1195,1196,1200],{},[193,1193,1194],{},"Economics of the loop."," Routing compact extract vs frontier judgement. Spend quotes. Seat pricing that includes unlimited flagship is an unengineered cost harness. ",[198,1197,1199],{"href":1198},"\u002Fblog\u002Fwhat-is-model-routing\u002F","Model routing",". Nimbus quotes each run; copilots meter seats. Different jobs.",[189,1202,1203,1206],{},[193,1204,1205],{},"Self-service vs programme."," If every new connector is a six-month SOW, you have bought a deployment, not a harness operators can tighten. That can still be the right buy for Ontology-scale complexity. It is the wrong buy for a standard Salesforce write this quarter.",[246,1208,1210],{"id":1209},"what-it-is-not","What it is not",[189,1212,1213,1214,298],{},"A coding harness with SSO. ",[198,1215,940],{"href":237},[189,1217,1218,1219,298],{},"A copilot with an admin console. ",[198,1220,1221],{"href":608},"Copilot vs work OS",[189,1223,1224,1225,1228,1229,298],{},"A framework. LangGraph can ",[438,1226,1227],{},"host"," an enterprise harness if you build grants, quotes, and a ledger. Out of the box it hosts a graph. ",[198,1230,1231],{"href":592},"Harness vs framework",[189,1233,1234],{},"“We integrate with Salesforce.” Integration is a slide. A scoped connector plus a blocked unsigned write is a harness.",[189,1236,1237,1238,495,1241,1244],{},"SWE-bench-first marketing. ",[198,1239,1240],{"href":280},"Anthropic",[198,1242,1243],{"href":224},"LangChain"," are writing about coding and general agents. Steal the discipline (stops, artifacts, sensors). Do not steal the benchmark as your control framework.",[246,1246,1248],{"id":1247},"thoughtworks-organisational-harness-in-operator-language","Thoughtworks’ organisational harness, in operator language",[189,1250,1251,1252,1255,1256,1259],{},"The ",[198,1253,1254],{"href":933},"Thoughtworks OS essay"," (10 July 2026) argues that most AI programmes fail because the organisation never built the operating system around the model: accountability, ownership, measurement, learning. They name four layers. An enterprise agent harness, as this article uses the term, is layers 3–4 made runnable for ",[438,1257,1258],{},"company jobs"," — not only for coding-agent users.",[189,1261,1262,1263,1266],{},"Delegation failures are the tell. The model was fine. The platform ran. Practitioner guides existed. The agent did what it was ",[438,1264,1265],{},"allowed"," to do. The company still took harm. Layer 4 questions: who approved that autonomy, who owns the policy, what was the escalation, how do we prevent the same miss on another team. Layers 1–3 cannot answer those. A chat product cannot either.",[189,1268,1269],{},"Thoughtworks’ control matrix is worth stealing even if you never hire them. Use deterministic controls where the boundary is knowable: allowed actions, residency, spend ceilings, blast-radius limits. Use probabilistic controls only where judgement is required. Pair every guide with a sensor. Temporal constraints — consistency across a multi-step workflow, not a single dropdown — are the ones they say teams miss most. A scheduling agent that is locally plausible on each step and globally inconsistent is not a “hallucination.” It is a missing temporal sensor.",[189,1271,1272],{},"Their public examples (Parloa’s repo-resident rules\u002Fskills\u002Fcommands; Morgan Stanley’s tiered autonomy on CVE triage) are coding-adjacent. Translate them: discount policy as a versioned wiki skill; “what delegation tier does this CRM write require?” instead of “do we trust the agent.” Nimbus’s Soft \u002F Hard \u002F Critical is that tiering in product form. AIP will have a different packaging. The architectural claim is the same.",[189,1274,1275,495,1277,1280],{},[198,1276,220],{"href":219},[198,1278,1279],{"href":212},"Wikipedia"," describe the runtime. Thoughtworks describe why a runtime without ownership still fails at scale. You need both descriptions when you buy.",[246,1282,1284],{"id":1283},"what-good-looks-like-on-a-live-job","What “good” looks like on a live job",[189,1286,1287],{},"A renewal write: workstream isolation; Salesforce attached read-only until write is enabled; wiki revision with the cap cited on the run; team cannot start if Legal’s connector requirement is missing; model proposes a quote; Hard gate; reject leaves Stage unchanged; export shows signer without a vendor screen-share. That is an enterprise harness. A demo that only answers “what should we do about Acme” is a copilot with a logo.",[189,1289,1290,1291,298],{},"Spend an hour asking where each Thoughtworks layer lives in the vendor’s product. If layer 4 is “our professional services team,” you are buying a programme. That can be the right buy. Name it. ",[198,1292,1293],{"href":1012},"Self-service vs FDE",[189,1295,1296,1297,1299,1300,1302],{},"Operators already know the human version of this harness. Maker-checker on journals. Segregation of duties on payments. Change-advisory on production. The enterprise agent harness is those instincts encoded so a model cannot talk through them. ",[198,1298,1086],{"href":1085}," did not wait for LLMs; it waited for a named signer. ",[198,1301,1081],{"href":1080}," did not wait for MCP; it waits for purpose limitation on the payload. If your AI programme cannot point to the interceptor that enforces those, you have a chatbot with a risk register.",[189,1304,1305],{},"What failure looks like in the first ninety days: every department clones a GPT with the same Salesforce key; Legal’s cap lives in a slide; the only eval is “the demo was impressive”; coding-agent MCP is pointed at production “just for a spike”; the ledger is Slack. What success looks like: one roster of teams, workstream isolation, default read, a Hard refuse on the first PoV, a wiki revision on the graph, inner harnesses still compiling in repos. Nimbus is built to make the success path a product week rather than a services year. Verify that claim with the refuse. AIP may be the right path when Ontology-scale complexity is real — then the harness is a programme, and you should staff it as one.",[246,1307,1309],{"id":1308},"how-this-shows-up-in-nimbus","How this shows up in Nimbus",[189,1311,1312,1313,1315,1316,1319,1320,1322,1323,1325],{},"Nimbus’s outer loop is: brief a ",[198,1314,543],{"href":372}," → assign a ",[198,1317,1318],{"href":368},"team"," whose connector contract is satisfied → retrieve under scope → draft on the canvas (Conflux) → quote writes → ",[198,1321,376],{"href":375}," pause → execute the signed payload → commit to the ",[198,1324,23],{"href":379},". Perception orients; it does not silently write. Routing picks model class per step.",[189,1327,1328,1329,1331],{},"That mapping is how we productised harness engineering for operators. It is not a claim that AIP or Agentforce are “not harnesses.” They are different time and scope. ",[198,1330,648],{"href":647}," is the scoring sheet for that runtime.",[246,1333,660],{"id":659},[662,1335,1337],{"id":1336},"do-we-need-this-if-we-already-have-claude-code","Do we need this if we already have Claude Code?",[189,1339,1340],{},"You need it for jobs whose workspace is the company. Keep Claude Code for repos. Do not share production SoR write tokens into the inner harness.",[662,1342,1344],{"id":1343},"is-palantir-aip-an-enterprise-harness","Is Palantir AIP an enterprise harness?",[189,1346,1347],{},"It can be, as a programme-shaped outer runtime. Ask deployment time, who sets a gate without vendor engineers, and whether the ledger is yours. Category yes; evaluation still required.",[662,1349,1351],{"id":1350},"is-agentforce-enough","Is Agentforce enough?",[189,1353,1354],{},"If the job is CRM-anchored and stays there, maybe. Cross-system jobs with Legal on the canvas usually need a harness that is not only Salesforce. Clear scopes; avoid two writers.",[662,1356,1358],{"id":1357},"can-we-build-this-on-langchain","Can we build this on LangChain?",[189,1360,1361,1362,1365],{},"Yes, with time. You will rebuild grants, quoting, roster, and replay. ",[198,1363,1364],{"href":1012},"Build vs buy",". Frameworks assemble loops; operators still need a loop they can hire.",[662,1367,1369],{"id":1368},"whats-the-first-proof","What’s the first proof?",[189,1371,1372,1373,1376],{},"A real cross-department write: operator attaches OAuth; unsigned payload blocked; reject leaves SoR unchanged; export shows signer. ",[198,1374,1375],{"href":647},"Proof of value",". A chat demo is not this.",[662,1378,719],{"id":718},[189,1380,1381,1383,1384,1383,1386,298],{},[198,1382,648],{"href":647},". ",[198,1385,728],{"href":727},[198,1387,233],{"href":232},[246,1389,736],{"id":735},[189,1391,1392,495,1394,298],{},[198,1393,748],{"href":356},[198,1395,1396],{"href":647},"how to evaluate an agent harness",[246,1398,752],{"id":751},[251,1400,1401,1406,1410,1414,1419,1425,1429,1433,1437,1441,1445,1449,1453,1458,1462,1467],{},[254,1402,1403],{},[198,1404,1405],{"href":200},"LangChain, Agents",[254,1407,1408],{},[198,1409,779],{"href":219},[254,1411,1412],{},[198,1413,774],{"href":212},[254,1415,1416],{},[198,1417,1418],{"href":933},"Thoughtworks, The operating system for enterprise AI",[254,1420,1421],{},[198,1422,1424],{"href":1423},"https:\u002F\u002Fwww.thoughtworks.com\u002Finsights\u002Fpodcasts\u002Ftechnology-podcasts\u002Fscaling-the-enterprise-harness--how-to-achieve-ai-agent-controll","Thoughtworks, Scaling the enterprise harness",[254,1426,1427],{},[198,1428,784],{"href":280},[254,1430,1431],{},[198,1432,804],{"href":389},[254,1434,1435],{},[198,1436,1097],{"href":429},[254,1438,1439],{},[198,1440,435],{"href":434},[254,1442,1443],{},[198,1444,1104],{"href":1103},[254,1446,1447],{},[198,1448,1091],{"href":1090},[254,1450,1451],{},[198,1452,1081],{"href":1080},[254,1454,1455],{},[198,1456,1457],{"href":1085},"SEC, Sarbanes–Oxley",[254,1459,1460],{},[198,1461,818],{"href":417},[254,1463,1464],{},[198,1465,1466],{"href":291},"Model Context Protocol specification",[254,1468,1469],{},[198,1470,1472],{"href":1471},"https:\u002F\u002Fwww.swebench.com\u002F","SWE-bench",{"title":169,"searchDepth":170,"depth":170,"links":1474},[1475,1476,1477,1478,1479,1480,1481,1482,1490,1491],{"id":248,"depth":170,"text":249},{"id":383,"depth":170,"text":384},{"id":1108,"depth":170,"text":1109},{"id":1209,"depth":170,"text":1210},{"id":1247,"depth":170,"text":1248},{"id":1283,"depth":170,"text":1284},{"id":1308,"depth":170,"text":1309},{"id":659,"depth":170,"text":660,"children":1483},[1484,1485,1486,1487,1488,1489],{"id":1336,"depth":849,"text":1337},{"id":1343,"depth":849,"text":1344},{"id":1350,"depth":849,"text":1351},{"id":1357,"depth":849,"text":1358},{"id":1368,"depth":849,"text":1369},{"id":718,"depth":849,"text":719},{"id":735,"depth":170,"text":736},{"id":751,"depth":170,"text":752},"An enterprise agent harness is the outer runtime for operators: a wiki, scoped connectors, agent teams, write gates, and a ledger.",{"eyebrow":861,"title":1494},"The outer runtime for operators",[1496,1499,1502],{"question":1497,"answer":1498},"What is an enterprise agent harness?","The outer runtime for operators: a wiki, scoped connectors, agent teams, write gates, and a ledger.",{"question":1500,"answer":1501},"Is a coding benchmark that harness?","No. A SWE-bench score does not grant a production login, and a chat with every production login is not a harness either.",{"question":1503,"answer":1504},"Who is it for?","Operators who need an agent to do company work and still be able to stop it.","\u002Fblog\u002Fwhat-is-an-enterprise-agent-harness",{"title":906,"description":1492},"blog\u002Fwhat-is-an-enterprise-agent-harness",[875,878,1509,376],"enterprise-ai","A1kbd3VxJotoXRDlHSlL2CNJkRzlKHNwzm2SnM8sYGc",{"id":1512,"title":1513,"archived":163,"authors":1514,"badge":1516,"body":1517,"date":858,"definedTerm":164,"department":164,"description":2102,"extension":172,"eyebrow":164,"faqHeader":2103,"faqs":2105,"footerBand":164,"headline":164,"image":164,"industry":164,"jobType":164,"listed":131,"location":164,"navigation":131,"openRoles":164,"pageLayout":164,"path":2115,"relatedHeading":164,"seo":2116,"series":875,"sitemap":131,"status":164,"stem":2117,"subhead":164,"tags":2118,"video":164,"whyJoin":164,"workplaceType":164,"__hash__":2120},"content\u002Fblog\u002Finner-vs-outer-agent-harness.md","Inner vs Outer Agent Harness",[1515],{"name":182,"to":183},{"label":185},{"type":166,"value":1518,"toc":2085},[1519,1532,1550,1567,1581,1583,1653,1674,1676,1679,1681,1695,1708,1711,1733,1741,1745,1762,1772,1782,1791,1802,1811,1826,1835,1848,1852,1855,1865,1873,1882,1886,1896,1906,1915,1920,1925,1933,1936,1938,1942,1945,1949,1956,1960,1970,1974,1979,1983,1990,1992,2005,2007,2013,2015],[189,1520,191,1521,1524,1525,1528,1529,1531],{},[193,1522,1523],{},"inner agent harness"," is the runtime around a model for a developer and a codebase. An ",[193,1526,1527],{},"outer agent harness"," is the runtime around a model for operators and live business systems. Same equation — ",[198,1530,205],{"href":200}," — different workspace, different sensors, different stop.",[189,1533,1534,1537,1538,1541,1542,1546,1547,298],{},[198,1535,1536],{"href":327},"Böckeler"," already uses “outer harness” for the controls ",[438,1539,1540],{},"users"," add around a coding agent (guides, sensors) as distinct from the vendor’s built-in loop. ",[198,1543,1545],{"href":1544},"https:\u002F\u002Faddyosmani.com\u002Fblog\u002Fown-the-outer-loop\u002F","Addy Osmani"," tells engineers to own the outer loop of investigate → implement → verify so accountability does not dissolve into the model. This article borrows those words and draws the cut enterprises actually buy: ",[193,1548,1549],{},"repo versus company",[189,1551,1552,1553,353,1556,626,1558,353,1560,1563,1564,1566],{},"Claude Code, Cursor, and Codex are excellent inner harnesses. They sandboxes, ",[311,1554,1555],{},"apply_patch",[311,1557,317],{},[311,1559,332],{},[198,1561,1562],{"href":307},"hooks",", and tests. Palantir AIP, Salesforce Agentforce, and OS-class products such as Nimbus are outer harnesses: ",[198,1565,352],{"href":351},", connectors, named signers, a decision record. Confusing them is how Legal is asked to “just use Cursor on the Salesforce repo” and how engineering is asked to “approve CRM writes in a coding agent.”",[189,1568,1569,1572,1573,1576,1577,1580],{},[198,1570,1571],{"href":612},"How to choose between a coding harness and an enterprise harness"," is the buying version of this page. ",[198,1574,1575],{"href":608},"Collaborative AI and personal assistants"," is the adjacent cut (personal assistant versus departmental work). Inner\u002Fouter is about ",[438,1578,1579],{},"which loop you are hiring",", not whether the UI looks like chat.",[246,1582,249],{"id":248},[251,1584,1585,1591,1597,1607,1619,1628,1640],{},[254,1586,1587,1590],{},[193,1588,1589],{},"Inner loop (classic SE)."," Edit, build, test on a developer’s machine. Fast. Local. The coding-agent inner harness lives here: shell, files, compiler.",[254,1592,1593,1596],{},[193,1594,1595],{},"Outer loop (classic SE)."," PR, CI, review, release. Osmani’s “own the outer loop” is this accountability layer for agentic coding. Still software.",[254,1598,1599,1602,1603,1606],{},[193,1600,1601],{},"Inner harness (this article)."," Vendor + user controls for a ",[193,1604,1605],{},"repository workspace",": Claude Code, Cursor, Codex. Eval: tests, Terminal-Bench, SWE-bench.",[254,1608,1609,1612,1613,1616,1617,298],{},[193,1610,1611],{},"Outer harness (this article)."," Controls for a ",[193,1614,1615],{},"company workspace",": jobs, systems of record, people who may sign. Eval: quoted write, identity, ledger. An ",[198,1618,243],{"href":242},[254,1620,1621,1624,1625,298],{},[193,1622,1623],{},"Guides vs sensors."," Feed-forward markdown versus feedback from tools. Inner: lint and pytest. Outer: schema of a Salesforce payload and a Hard gate. See ",[198,1626,1627],{"href":232},"what is harness engineering",[254,1629,1630,1633,1634,1637,1638,298],{},[193,1631,1632],{},"CLAUDE.md \u002F AGENTS.md."," Inner guides. ",[198,1635,1240],{"href":1636},"https:\u002F\u002Fclaude.com\u002Fblog\u002Fsteering-claude-code-skills-hooks-rules-subagents-and-more"," is explicit: files are context; hooks are deterministic. A company wiki is the outer analogue of those files — asserted policy, not a repo README. See ",[198,1639,499],{"href":498},[254,1641,1642,1645,1646,1649,1650,1652],{},[193,1643,1644],{},"Write gate."," Inner: hook denies ",[311,1647,1648],{},"rm"," or force-push. Outer: ",[198,1651,511],{"href":356}," — adapter cannot mutate CRM until a named role signs the quote.",[189,1654,1655,1656,1658,1659,353,1661,1664,1665,1667,1668,1658,1670,1673],{},"Nimbus is built as an outer harness: ",[198,1657,365],{"href":364}," instead of only ",[311,1660,332],{},[198,1662,1663],{"href":1142},"connectors"," instead of only a local shell, ",[198,1666,376],{"href":375}," instead of only a pre-commit hook, ",[198,1669,23],{"href":379},[311,1671,1672],{},"git log",". Engineering should still run Claude Code. Those products should not share a write path to NetSuite.",[246,1675,384],{"id":383},[189,1677,1678],{},"Demos collapse the cut. Both products answer a question. Both call tools. Both show a transcript. The evaluation is the workspace.",[189,1680,394],{},[251,1682,1683,1686,1689,1692],{},[254,1684,1685],{},"Security asks whether the coding agent’s MCP server can reach production Salesforce",[254,1687,1688],{},"RevOps wants “an agent” and is shown a SWE-bench slide",[254,1690,1691],{},"Engineering wants Cursor and is told to wait for the enterprise OS",[254,1693,1694],{},"You already have both, and they silently write to the same object",[189,1696,1697,1699,1700,1704,1705,1707],{},[198,1698,1037],{"href":389}," describes agentic systems as an organisational design problem. Inner harnesses scale developer throughput. They do not, by themselves, scale governed operations. ",[198,1701,1703],{"href":1702},"https:\u002F\u002Fhai.stanford.edu\u002Fai-index\u002F2025-ai-index-report","Stanford HAI’s 2025 AI Index"," maps how fast coding-agent tooling moved. Speed in the repo is not a substitute for ",[198,1706,1091],{"href":1090}," oversight on systems that affect customers and money.",[189,1709,1710],{},"Two failure modes:",[1712,1713,1714,1724],"ol",{},[254,1715,1716,1719,1720,1723],{},[193,1717,1718],{},"Outer job, inner harness."," A pricing change drafted in Cursor with an MCP Salesforce tool. Tests pass on a fixture. Production Amount changes. ",[311,1721,1722],{},"git blame"," does not name the signer. You used a repo loop on a company record.",[254,1725,1726,1729,1730,1732],{},[193,1727,1728],{},"Inner job, outer harness."," “Rewrite this function” opened as a cross-department ",[198,1731,543],{"href":372}," with a Critical gate. Engineers will route around it. You used a company loop on a compile.",[189,1734,1735,1737,1738,1740],{},[198,1736,1097],{"href":429}," Map step: know the context of use. Inner and outer are different contexts. ",[198,1739,435],{"href":434}," wants controls matched to that context. One harness policy for “all AI” is how both jobs get the wrong stop.",[246,1742,1744],{"id":1743},"what-each-harness-actually-owns","What each harness actually owns",[189,1746,1747,1750,1751,1754,1755,1758,1759,1761],{},[193,1748,1749],{},"Workspace."," Inner: a checkout, often sandboxed. Anthropic’s ",[198,1752,1753],{"href":464},"long-running harness"," keeps progress in git and files because the workspace ",[438,1756,1757],{},"is"," the filesystem. Outer: a job folder with people, budget, and attached systems — a ",[198,1760,543],{"href":351},". Files may appear as artefacts. They are not the system of record.",[189,1763,1764,1767,1768,1771],{},[193,1765,1766],{},"Identity."," Inner: the developer’s machine credentials, a repo token, maybe a sandbox role. Outer: org roster, workstream membership, named approver. The model is not the principal. ",[198,1769,1770],{"href":1142},"Connector and permissions architecture"," is the outer identity plane.",[189,1773,1774,1777,1778,1781],{},[193,1775,1776],{},"Tools."," Inner: shell, editor, tests, browser, maybe MCP to docs. Outer: CRM, ERP, warehouse, ticket systems, mail — default read, write as a separate plane. ",[198,1779,1780],{"href":296},"MCP"," can sit under both. The grant must not.",[189,1783,1784,1787,1788,1790],{},[193,1785,1786],{},"Guides."," Inner: ",[311,1789,332],{},", skills, directory-local rules. Outer: company wiki, playbooks versioned with the run. Mixing them is useful (engineering conventions in the repo; discount policy in the wiki). Collapsing them is how a style guide becomes “legal approval.”",[189,1792,1793,1796,1797,1801],{},[193,1794,1795],{},"Sensors."," Inner: typechecker, unit tests, CI, architecture tests. Böckeler and ",[198,1798,1800],{"href":1799},"https:\u002F\u002Fwww.thoughtworks.com\u002Fen-us\u002Finsights\u002Fblog\u002Fgenerative-ai\u002Fharness-engineering-agent-feedback-exploring-ai-coding-sensors","Thoughtworks on sensors",". Outer: payload schema, blast-radius cardinality, maker-checker, exportable ledger. A passing pytest does not mean Opportunity.Stage was authorised.",[189,1803,1804,1806,1807,1810],{},[193,1805,276],{}," Inner: tests red, hook exit 2, max steps, human in the IDE. Outer: wait-for-named-signer, missing connector, budget, reject. ",[198,1808,1809],{"href":1166},"Human-in-the-loop"," in a coding agent is “the developer kept going.” HITL in an outer harness is a first-class step with identity.",[189,1812,1813,1787,1816,353,1818,1822,1823,1825],{},[193,1814,1815],{},"Eval.",[198,1817,1472],{"href":1471},[198,1819,1821],{"href":1820},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2601.11868","Terminal-Bench",", your suite. Outer: replay the signer; compare quote to SoR; see ",[198,1824,1189],{"href":521},". Leaderboard scores are not a SOX control.",[189,1827,1828,1831,1832,1834],{},[193,1829,1830],{},"Memory."," Inner: files, commits, session transcripts, memory files the next coding session loads. Outer: wiki + ",[198,1833,23],{"href":493}," so next quarter’s operator can ask why a field changed. Chat logs of a coding session are not institutional memory for RevOps.",[189,1836,1837,1838,1840,1841,1843,1844,1847],{},"Nimbus’s ",[198,1839,369],{"href":368}," sit on the outer side: mandates, required connectors, approval triggers. You can still ",[438,1842,1041],{}," an inner harness as a bounded tool behind a connector (for example a coding agent that only opens a draft PR). Do not let that inner harness become the orchestrator of record for a CRM write. ",[198,1845,1846],{"href":531},"Multi-agent architecture"," says the same thing with specialists: hands are not roles.",[246,1849,1851],{"id":1850},"how-they-should-sit-together","How they should sit together",[189,1853,1854],{},"Most companies need both. That is not a hedge. It is how software and operations already split.",[189,1856,1857,1860,1861,1864],{},[193,1858,1859],{},"Pattern that works."," Engineers use Cursor or Claude Code on application repos. CI remains the merge sensor. Separately, RevOps and Finance run outer-harness jobs on Salesforce and NetSuite. If a coding agent must touch a live business system, it proposes an artefact; the outer harness quotes and gates the write. Two writers to the same object without a single quote is the failure ",[198,1862,1863],{"href":531},"multi-agent architecture"," already names.",[189,1866,1867,1870,1871,298],{},[193,1868,1869],{},"Pattern that fails."," One MCP mesh with production tokens, used from the IDE and from the chatbot and from the OS. Confused deputy. ",[198,1872,297],{"href":296},[189,1874,1875,1878,1879,298],{},[193,1876,1877],{},"Thoughtworks’ four layers"," — model, builder harness, user harness, organisational harness — map cleanly: Claude Code is builder + user on the inner side; the organisational layer is the outer operating model. Nimbus is one productisation of that outer layer, not the only one. AIP is a programme-shaped outer harness. Agentforce is CRM-anchored. Score scope and time-to-value separately. See ",[198,1880,1881],{"href":1012},"self-service vs forward-deployed",[246,1883,1885],{"id":1884},"a-week-that-uses-both","A week that uses both",[189,1887,1888,1889,1891,1892,1895],{},"Monday an engineer uses Cursor to fix a pricing calculator in the billing service. ",[311,1890,332],{}," says no raw SQL in the request path. A hook blocks ",[311,1893,1894],{},"git push --force",". CI runs the unit suite. The PR is the artefact. CODEOWNERS signs the merge. That is a complete inner story. SWE-bench is relevant only as a vendor quality signal for the coding tool, not as a control.",[189,1897,1898,1899,1901,1902,1905],{},"Tuesday RevOps needs the list price on twenty renewals updated after Legal changed the cap in the playbook. The artefact is Salesforce. The signer is a named RevOps lead. The sensor is: quoted fields, hash, read-back. If Tuesday’s job is opened as a Cursor session with an MCP Salesforce server using a shared integration user, you have imported Monday’s workspace into Tuesday’s system of record. ",[311,1900,1672],{}," will not name the RevOps lead. ",[198,1903,1904],{"href":356},"Write-back"," did not fire because the inner harness does not have that interceptor.",[189,1907,1908,1909,1911,1912,1914],{},"Wednesday someone proposes “one agent for everything.” The honest architecture is: Monday’s harness stays. Tuesday’s job runs on an outer harness — in Nimbus, a ",[198,1910,543],{"href":372}," with the CRM connector, the wiki revision that contains the new cap, a Hard gate. If the calculator ",[438,1913,311],{}," must change as well, the outer job can spawn a bounded inner step that opens a draft PR. Two artefacts, two sensors, one company rule: unsigned SoR writes are impossible from either loop.",[189,1916,1917,1918,298],{},"Thursday Security reviews MCP. The question is not “is MCP approved.” It is “which workspace may this server mutate.” Inner: sandbox and repo. Outer: workstream grant. Same protocol, different identity box. ",[198,1919,297],{"href":296},[189,1921,1922,1923,298],{},"Friday you look at evals. Engineering posts a Terminal-Bench plot for the coding vendor. Finance asks who signed Amount. Those are not competing dashboards. They are different oracles. ",[198,1924,1003],{"href":521},[189,1926,1927,1929,1930,1932],{},[198,1928,934],{"href":933}," would call Monday layers 2–3 on a builder harness, Tuesday a delegation question on layer 4, and “one agent” a way to skip layer 4. ",[198,1931,703],{"href":1544}," would say engineering still owns verify-and-merge on Monday. Neither author is selling Nimbus. Both are describing why the cut exists.",[189,1934,1935],{},"If you only fund inner harnesses, Tuesday happens in paste and Slack. If you only fund outer harnesses, Monday happens in unsanctioned Cursor anyway. Fund both. Bind writes.",[246,1937,660],{"id":659},[662,1939,1941],{"id":1940},"is-cursor-an-enterprise-harness-if-we-sso-it","Is Cursor an enterprise harness if we SSO it?",[189,1943,1944],{},"SSO is admin control. It does not quote a NetSuite journal or bind a Finance signer. Cursor can be an inner harness in an enterprise. That is not the same as an outer harness.",[662,1946,1948],{"id":1947},"can-claude-code-hooks-replace-write-back-governance","Can Claude Code hooks replace write-back governance?",[189,1950,1951,1952,1955],{},"They can replace ",[438,1953,1954],{},"some"," inner invariants (dangerous bash). They do not give you a payload in the language of Salesforce, a roster-aware approver, or an exportable operations ledger. Different workspace.",[662,1957,1959],{"id":1958},"should-we-ban-coding-agents-until-the-os-is-live","Should we ban coding agents until the OS is live?",[189,1961,1962,1963,1966,1967,298],{},"Usually no. Ban unsigned writes to systems of record from ",[438,1964,1965],{},"any"," agent, inner or outer. Let inner harnesses keep compiling. ",[198,1968,1969],{"href":356},"How to solve unapproved CRM writes from AI",[662,1971,1973],{"id":1972},"where-does-a-copilot-fit","Where does a copilot fit?",[189,1975,1976,1977,298],{},"A copilot is often not a full inner harness — no repo loop, no tests. Personal throughput. Keep it for mail. Do not give it the CRM write token. ",[198,1978,1221],{"href":608},[662,1980,1982],{"id":1981},"is-nimbus-trying-to-replace-claude-code","Is Nimbus trying to replace Claude Code?",[189,1984,1985,1986,1989],{},"No. Different workspace. Nimbus is the company loop; Claude Code is the repo loop. ",[198,1987,11],{"href":1988},"\u002Foverview\u002F"," is the product map. This page is the architectural cut.",[662,1991,719],{"id":718},[189,1993,1994,1383,1997,2000,2001,2004],{},[198,1995,1996],{"href":242},"What is an enterprise agent harness",[198,1998,1999],{"href":592},"Agent harness vs agent framework"," if you are assembling rather than hiring. ",[198,2002,2003],{"href":1115},"What is an agent harness"," for the base noun.",[246,2006,736],{"id":735},[189,2008,2009,495,2011,298],{},[198,2010,233],{"href":232},[198,2012,728],{"href":727},[246,2014,752],{"id":751},[251,2016,2017,2021,2025,2030,2034,2038,2042,2047,2051,2055,2060,2064,2069,2073,2077,2081],{},[254,2018,2019],{},[198,2020,1405],{"href":200},[254,2022,2023],{},[198,2024,794],{"href":327},[254,2026,2027],{},[198,2028,2029],{"href":1544},"Addy Osmani, Own the outer loop",[254,2031,2032],{},[198,2033,799],{"href":446},[254,2035,2036],{},[198,2037,1418],{"href":933},[254,2039,2040],{},[198,2041,789],{"href":464},[254,2043,2044],{},[198,2045,2046],{"href":1636},"Anthropic, Steering Claude Code",[254,2048,2049],{},[198,2050,838],{"href":307},[254,2052,2053],{},[198,2054,1472],{"href":1471},[254,2056,2057],{},[198,2058,2059],{"href":1820},"Terminal-Bench (arXiv:2601.11868)",[254,2061,2062],{},[198,2063,804],{"href":389},[254,2065,2066],{},[198,2067,2068],{"href":1702},"Stanford HAI, 2025 AI Index",[254,2070,2071],{},[198,2072,1097],{"href":429},[254,2074,2075],{},[198,2076,435],{"href":434},[254,2078,2079],{},[198,2080,1091],{"href":1090},[254,2082,2083],{},[198,2084,1466],{"href":291},{"title":169,"searchDepth":170,"depth":170,"links":2086},[2087,2088,2089,2090,2091,2092,2100,2101],{"id":248,"depth":170,"text":249},{"id":383,"depth":170,"text":384},{"id":1743,"depth":170,"text":1744},{"id":1850,"depth":170,"text":1851},{"id":1884,"depth":170,"text":1885},{"id":659,"depth":170,"text":660,"children":2093},[2094,2095,2096,2097,2098,2099],{"id":1940,"depth":849,"text":1941},{"id":1947,"depth":849,"text":1948},{"id":1958,"depth":849,"text":1959},{"id":1972,"depth":849,"text":1973},{"id":1981,"depth":849,"text":1982},{"id":718,"depth":849,"text":719},{"id":735,"depth":170,"text":736},{"id":751,"depth":170,"text":752},"An inner agent harness runs a developer and a repository — CLAUDE.md, hooks, tests.",{"eyebrow":861,"title":2104},"The repo and the company",[2106,2109,2112],{"question":2107,"answer":2108},"What is an inner agent harness?","The runtime for a developer and a repository: CLAUDE.md, hooks, and tests.",{"question":2110,"answer":2111},"What is an outer harness?","The runtime for the company: wiki, connectors, write gates, and a ledger of what changed.",{"question":2113,"answer":2114},"Can one replace the other?","No. Most enterprises need both. A perfect repo harness does not stop an unapproved CRM write, and a company gate does not run the unit tests.","\u002Fblog\u002Finner-vs-outer-agent-harness",{"title":1513,"description":2102},"blog\u002Finner-vs-outer-agent-harness",[875,878,2119,1509],"coding-agents","03H2FWvQKQ3zCeocBdx6w-GVtDt-Nz1nBE0Sz22oZ2I",{"enabled":163,"message":2122,"linkLabel":79,"linkHref":80,"id":2123,"title":2124,"archived":163,"authors":164,"badge":164,"body":2125,"date":164,"definedTerm":164,"department":164,"description":169,"extension":172,"eyebrow":164,"faqHeader":164,"faqs":164,"footerBand":164,"headline":164,"image":164,"industry":164,"jobType":164,"listed":131,"location":164,"navigation":131,"openRoles":164,"pageLayout":164,"path":2129,"relatedHeading":164,"seo":2130,"series":164,"sitemap":163,"status":164,"stem":2131,"subhead":164,"tags":164,"video":164,"whyJoin":164,"workplaceType":164,"__hash__":2132},"We're hiring! Join the team building the Sentient Enterprise.","content\u002Fshared\u002Fhiring.md","Hiring banner",{"type":166,"value":2126,"toc":2127},[],{"title":169,"searchDepth":170,"depth":170,"links":2128},[],"\u002Fshared\u002Fhiring",{"title":2124,"description":169},"shared\u002Fhiring","1zs3boivKda1e-b-hAyuNcmZSKjZUAXmecnwHVgcHzk",{"fold":2134,"id":2138,"title":2139,"archived":163,"authors":164,"badge":164,"body":2140,"date":164,"definedTerm":164,"department":164,"description":169,"extension":172,"eyebrow":164,"faqHeader":164,"faqs":164,"footerBand":2144,"headline":164,"image":164,"industry":164,"jobType":164,"listed":131,"location":164,"navigation":131,"openRoles":164,"pageLayout":164,"path":2148,"relatedHeading":164,"seo":2149,"series":164,"sitemap":163,"status":164,"stem":2150,"subhead":164,"tags":164,"video":164,"whyJoin":164,"workplaceType":164,"__hash__":2151},{"headline":2135,"description":2136,"primaryLabel":8,"primaryTo":2137,"secondaryLabel":898,"secondaryTo":12},"Run frontier AI your business actually owns.","Governed workstreams, 3,000+ integrations, and a proprietary knowledge graph. Start on Free.","\u002Fsignup?plan=free","content\u002Fshared\u002Fcta.md","Site CTAs",{"type":166,"value":2141,"toc":2142},[],{"title":169,"searchDepth":170,"depth":170,"links":2143},[],{"headline":2145,"description":2146,"primaryLabel":8,"primaryTo":2137,"secondaryLabel":2147,"secondaryTo":85},"See what governed AI looks like on your stack.","Connect your tools, run a workstream, and keep every decision on your ledger. Start on Free.","Talk to our team","\u002Fshared\u002Fcta",{"title":2139,"description":169},"shared\u002Fcta","eYqahyaPnbp8GKrWpoORbZdtmkWmgHr5F61ZHOnb8sY",1792999485551]