[{"data":1,"prerenderedAt":1570},["ShallowReactive",2],{"site-nav-content":3,"blog:/blog/what-is-multi-agent-ai":163,"blog-index-copy":650,"blog:/blog/what-is-multi-agent-ai:surround":671,"hiring-banner-content":1539,"site-cta-content":1551},{"header":4,"productNav":9,"nav":42,"footer":61,"askAI":116,"id":147,"title":148,"archived":149,"authors":150,"badge":150,"body":151,"date":150,"department":150,"description":155,"extension":158,"eyebrow":150,"faqHeader":150,"faqs":150,"footerBand":150,"headline":150,"image":150,"industry":150,"jobType":150,"listed":149,"location":150,"navigation":115,"openRoles":150,"pageLayout":150,"path":159,"relatedHeading":150,"seo":160,"series":150,"sitemap":115,"status":150,"stem":161,"subhead":150,"tags":150,"video":150,"whyJoin":150,"workplaceType":150,"__hash__":162},{"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":85,"socialLinks":95,"bottomLinks":105},"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.",[71,72,73,74,75,77,80,83],{"label":47,"to":48},{"label":50,"to":51},{"label":53,"to":54},{"label":59,"to":60},{"label":76,"to":57},"Partner Program",{"label":78,"to":79},"Careers","/careers",{"label":81,"to":82},"System status","/status",{"label":7,"to":84},"/contact",[86,89,92],{"label":87,"to":88},"Terms of Service","/terms",{"label":90,"to":91},"Privacy Policy","/privacy",{"label":93,"to":94},"Compliance","/compliance",[96,99,102],{"label":97,"href":98},"LinkedIn","https://www.linkedin.com/company/gonimbusai/",{"label":100,"href":101},"X","https://x.com/gonimbusai",{"label":103,"href":104},"Instagram","https://www.instagram.com/gonimbus_ai/",[106,108,110,111,112],{"label":107,"to":88},"Terms",{"label":109,"to":91},"Privacy",{"label":93,"to":94},{"label":81,"to":82},{"label":113,"to":114,"external":115},"LLMs.txt","/llms.txt",true,{"text":117,"prompt":118},"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":119,"platforms":121},{" Summarize the highlights from Nimbus's website":120},"https://gonimbus.ai",[122,127,132,137,142],{"name":123,"label":124,"icon":125,"hrefPrefix":126},"chatgpt","ChatGPT","simple-icons:openai","https://chatgpt.com/?prompt=",{"name":128,"label":129,"icon":130,"hrefPrefix":131},"perplexity","Perplexity","mdi:magnify","https://www.perplexity.ai/search/new?q=",{"name":133,"label":134,"icon":135,"hrefPrefix":136},"grok","Grok","simple-icons:x","https://x.com/i/grok?text=",{"name":138,"label":139,"icon":140,"hrefPrefix":141},"claude","Claude","simple-icons:anthropic","https://claude.ai/new?q=",{"name":143,"label":144,"icon":145,"hrefPrefix":146},"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":152,"value":153,"toc":154},"minimark",[],{"title":155,"searchDepth":156,"depth":156,"links":157},"",2,[],"md","/shared/nav",{"title":148,"description":155},"shared/nav","p6IsjEfjcrwkjwvGsPcuEkpoJSeUn2vJXPTxLWQfw9M",{"id":164,"title":165,"archived":149,"authors":166,"badge":169,"body":171,"date":639,"department":150,"description":640,"extension":158,"eyebrow":150,"faqHeader":150,"faqs":150,"footerBand":150,"headline":150,"image":150,"industry":150,"jobType":150,"listed":149,"location":150,"navigation":115,"openRoles":150,"pageLayout":150,"path":641,"relatedHeading":150,"seo":642,"series":643,"sitemap":115,"status":150,"stem":644,"subhead":150,"tags":645,"video":150,"whyJoin":150,"workplaceType":150,"__hash__":649},"content/blog/what-is-multi-agent-ai.md","What is Multi-Agent AI",[167],{"name":168,"to":120},"Nimbus Research",{"label":170},"Explainer",{"type":152,"value":172,"toc":614},[173,177,180,196,201,264,280,288,292,295,298,309,312,333,340,346,351,357,368,374,380,386,390,396,402,408,414,420,426,429,435,451,457,463,467,477,480,493,497,501,504,508,511,515,521,525,528,532,535,539,542,546,553,557,560,564,567,571,576,580,583,587,590,594,602,606],[174,175,176],"p",{},"Multi-agent AI is more than one AI specialist handing work to each other, the way legal already reviews a go-to-market draft. They share a goal, pass intermediate work, and stop for a human when duties collide.",[174,178,179],{},"It is not “more than one model call.” If you can rename your agents to “saved prompts” and nothing breaks, you do not have multi-agent AI. You have prompt folders. Three chat tabs labelled Research, CRM, and Legal are still one person copying between windows.",[174,181,182,183,187,188,195],{},"Companies already hand work between departments. Multi-agent AI is useful when those hand-offs ",[184,185,186],"em",{},"are"," the job. It is not useful as a prestige multiplier on a task one specialist should finish. ",[189,190,194],"a",{"href":191,"rel":192},"https://www.anthropic.com/engineering/building-effective-agents",[193],"nofollow","Anthropic’s note on building effective agents"," is mostly about workflows and stops, not about collecting a zoo of bots.",[197,198,200],"h2",{"id":199},"words-youll-hear","Words you’ll hear",[202,203,204,212,218,224,230,236,247,253],"ul",{},[205,206,207,211],"li",{},[208,209,210],"strong",{},"Single-agent."," One policy, one tool set, one conversation. The human is the only coordinator. Right for many tasks: rewrite this email, explain this clause.",[205,213,214,217],{},[208,215,216],{},"Multi-agent."," Role specialisation, shared state scoped to the job, arbitration when agents disagree, and a stop — including “wait for approval.”",[205,219,220,223],{},[208,221,222],{},"Orchestrator."," A coordinator that assigns work to specialists. Not a licence to give every specialist the same production login.",[205,225,226,229],{},[208,227,228],{},"Agent team."," Nimbus’s product name for a department-shaped specialist group (finance, revenue, operations) that persists, rather than a zoo of user-owned bots.",[205,231,232,235],{},[208,233,234],{},"Separation of duties."," The specialist that recommends a CRM update is not the same principal that executes it without a quote.",[205,237,238,241,242,246],{},[208,239,240],{},"Shared state."," The job’s brief, wiki sections, and artefacts — not a pile of private chats. At work, this is the ",[189,243,245],{"href":244},"what-is-an-ai-workstream","workstream",".",[205,248,249,252],{},[208,250,251],{},"Arbitration."," What happens when specialists disagree. At work, legal’s “do not send” should beat go-to-market’s “looks fine,” and a named human still releases.",[205,254,255,258,259,263],{},[208,256,257],{},"Cast versus brain."," Several agents is a cast. ",[189,260,262],{"href":261},"what-is-model-routing","Model routing"," is which brain each step pays for.",[174,265,266,267,270,271,274,275,279],{},"Tools are ",[208,268,269],{},"hands",". They are not ",[208,272,273],{},"roles",". A common plug so AI apps can use the same tools is useful plumbing — see ",[189,276,278],{"href":277},"what-is-model-context-protocol","What is Model Context Protocol"," — and it is also how a “multi-agent” demo quietly becomes one actor with every tool on the belt.",[174,281,282,283,287],{},"An ",[189,284,286],{"href":285},"what-is-an-agentic-workflow","agentic workflow"," is the sequence. Multi-agent AI is the cast. Mixing those words is how vendors sell extra model calls as organisation design.",[197,289,291],{"id":290},"why-you-should-care","Why you should care",[174,293,294],{},"Coordination is an org-chart problem, not a model problem.",[174,296,297],{},"It affects you if:",[202,299,300,303,306],{},[205,301,302],{},"one “god agent” would need every production login",[205,304,305],{},"legal must review a draft before anyone writes CRM",[205,307,308],{},"next quarter, “why did we change this?” must still be answerable",[174,310,311],{},"Known failure modes:",[313,314,315,321,327],"ol",{},[205,316,317,320],{},[208,318,319],{},"Parallel single-agent."," Three chat tabs. No shared state. The user is the message bus.",[205,322,323,326],{},[208,324,325],{},"Agent sprawl."," Dozens of custom agents with overlapping tools and unclear write rights.",[205,328,329,332],{},[208,330,331],{},"Orchestration without memory."," A beautiful run that discards the outcome when the worker exits.",[174,334,335,336,246],{},"Multi-agent does not reduce accountability. It concentrates it on the release gate. See ",[189,337,339],{"href":338},"what-is-human-in-the-loop-ai","What is human-in-the-loop AI",[174,341,342,343,246],{},"Do not think in “number of agents.” Think in ",[208,344,345],{},"jobs that already have hand-offs",[347,348,350],"h3",{"id":349},"what-changes-by-role","What changes by role",[174,352,353,356],{},[208,354,355],{},"Finance."," A finance-shaped agent team can draft a journal against the close checklist without inheriting GTM’s CRM write connector. Separation of duties is the point. Finance should still be the named signer on the ledger. Extra agents are not extra authorisation.",[174,358,359,362,363,367],{},[208,360,361],{},"Legal."," Review-before-send is a real hand-off. Legal-shaped specialists should not need People Ops files “for context.” Legal also cares that internal agent debate is not treated as a signature. The ",[189,364,366],{"href":365},"what-is-a-lifecycle-graph","lifecycle graph"," should show the human at release.",[174,369,370,373],{},[208,371,372],{},"Operations."," Persist teams, do not spawn a bot per user. Ops should refuse sprawl, insist on shared workstream state, and keep fail-closed writes outside the cast. Incident reviews need one chain, not three private transcripts.",[174,375,376,379],{},[208,377,378],{},"Go-to-market."," Cross-functional launches already look like this: GTM drafts, legal redlines, finance checks the discount. Encode that. Do not encode a god agent that can do all three logins. Time-to-approved-write still beats number-of-agents as a metric.",[174,381,382,385],{},[208,383,384],{},"Security."," Sprawl is an identity problem. Each specialist with overlapping write tools is another path to production. Prompt injection that turns one specialist into a tool-caller should still die at the gate. Least privilege applies per role, not “the swarm is trusted.”",[347,387,389],{"id":388},"what-people-get-wrong","What people get wrong",[174,391,392,395],{},[208,393,394],{},"Saved prompts as agents."," If renaming them changes nothing, they were prompts.",[174,397,398,401],{},[208,399,400],{},"Chat tabs as multi-agent."," The user is still the bus.",[174,403,404,407],{},[208,405,406],{},"More agents as more quality."," Coordination cost is real. Start from existing hand-offs.",[174,409,410,413],{},[208,411,412],{},"Agents as signers."," Internal critics reduce garbage. They are not the named human.",[174,415,416,419],{},[208,417,418],{},"Every specialist gets every tool."," That recreates the god agent with extra steps.",[174,421,422,425],{},[208,423,424],{},"Orchestration without a workstream."," No scope, no budget, no memory.",[174,427,428],{},"Good looks like: department-shaped teams that persist, inherit workstream scope (wiki, connectors, NTU budget), disagree in the open, and stop for a named signer. Failure looks like a folder of user-owned bots and a demo where five helpers share one production key.",[174,430,431,432,434],{},"A useful test: draw the analogue hand-off first. If legal already reviews a go-to-market draft before a customer sees it, you have a candidate for two specialist roles on one ",[189,433,245],{"href":244},". If one analyst extracts a table, you have a candidate for a single tool-using agent. If nobody can name the hand-off, you are inventing a cast for a play that does not exist — and you will invent overlapping tools to keep them busy.",[174,436,437,438,441,442,446,447,450],{},"Spend follows the same test. Extra specialists mean extra model calls. Without ",[189,439,440],{"href":261},"routing"," and an ",[189,443,445],{"href":444},"what-is-ai-token-economics","NTU"," ceiling, “let them debate” is an unbounded loop. Debate that never reaches a named signer is also not ",[189,448,449],{"href":338},"human-in-the-loop","; it is theatre with more speakers.",[174,452,453,454,456],{},"Memory is the other test. If the hand-off is not on the ",[189,455,366],{"href":365},", next quarter’s question — “why did we change this?” — has no answer except whoever still remembers the swarm. That is not multi-agent AI. That is parallel chat.",[174,458,459,462],{},[189,460,461],{"href":20},"Agent teams"," in Nimbus are meant to look like the departments you already have, not like a prompt gallery. If your org chart does not contain a role, do not invent an agent for it. If your org chart does contain a role that must review before release, do not skip it because a single flagship model offered to “do it all.” Number of agents is a vanity metric. Named hand-offs are not.",[197,464,466],{"id":465},"how-this-shows-up-in-nimbus","How this shows up in Nimbus",[174,468,469,470,473,474,476],{},"Nimbus implements multi-agent AI as ",[208,471,472],{},"agent teams",", not as a folder of user-owned bots. Teams persist. They inherit ",[189,475,245],{"href":244}," scope — wiki sections, connectors, spend budget — and they participate in the same release process as any other actor.",[174,478,479],{},"Connectors stay read-only by default. Agent teams can draft. They cannot waive the gate. The Lifecycle Graph records the hand-offs as work, not as a swarm mystery.",[174,481,482,483,485,486,488,489,246],{},"See ",[189,484,461],{"href":20},". An ",[189,487,286],{"href":285}," is the sequence. Multi-agent AI is the cast. The OS around them is ",[189,490,492],{"href":491},"what-is-an-enterprise-ai-operating-system","What is an enterprise AI operating system",[197,494,496],{"id":495},"questions-people-actually-ask","Questions people actually ask",[347,498,500],{"id":499},"isnt-this-just-several-chatgpts-talking","Isn’t this just several ChatGPTs talking?",[174,502,503],{},"Not if they share one job, one scope, and one stop. Several chats with no shared state is still you, copying.",[347,505,507],{"id":506},"do-we-need-multi-agent-ai-for-everything","Do we need multi-agent AI for everything?",[174,509,510],{},"No. A single tool-using agent is enough for many tasks. Add specialists when duties already split in the organisation.",[347,512,514],{"id":513},"does-each-agent-need-its-own-model","Does each agent need its own model?",[174,516,517,518,246],{},"Often yes for cost and quality. Classification rarely needs the flagship. Tricky policy interpretation often does. See ",[189,519,520],{"href":261},"What is model routing",[347,522,524],{"id":523},"who-is-accountable-when-several-agents-worked-on-it","Who is accountable when several agents worked on it?",[174,526,527],{},"The named human at release — not “the swarm.” Internal critics can reduce garbage that reaches the person. They are not the signer.",[347,529,531],{"id":530},"how-is-this-different-from-an-agentic-workflow","How is this different from an agentic workflow?",[174,533,534],{},"The workflow is the sequence of steps and stops. Multi-agent is whether more than one specialist role executes those steps. You can have a workflow with one agent.",[347,536,538],{"id":537},"what-is-an-agent-team-in-nimbus","What is an agent team in Nimbus?",[174,540,541],{},"A department-shaped specialist group that persists and inherits the workstream’s wiki, connectors, and budget — not a user-owned custom GPT.",[347,543,545],{"id":544},"can-agents-approve-each-others-writes","Can agents approve each other’s writes?",[174,547,548,549,246],{},"They can flag problems. Execution still needs a named signer and a fail-closed gate. See ",[189,550,552],{"href":551},"what-is-write-back-governance","What is write-back governance",[347,554,556],{"id":555},"why-not-one-god-agent-with-every-connector","Why not one god agent with every connector?",[174,558,559],{},"Because least privilege and separation of duties already exist in the company. Encoding the org chart is safer than encoding a superuser.",[347,561,563],{"id":562},"how-do-we-avoid-agent-sprawl","How do we avoid agent sprawl?",[174,565,566],{},"One team per function that already exists, assigned onto jobs, with overlapping tools treated as an incident. Do not let every operator publish a bot.",[347,568,570],{"id":569},"does-mcp-make-us-multi-agent","Does MCP make us multi-agent?",[174,572,573,574,246],{},"No. MCP is how a host calls tools. Many helpers on one belt can still be one actor. See ",[189,575,278],{"href":277},[347,577,579],{"id":578},"how-do-disagreements-get-recorded","How do disagreements get recorded?",[174,581,582],{},"On the decision chain: what was proposed, what was objected to, what the human signed. If disagreement evaporates with the session, you have orchestration without memory.",[347,584,586],{"id":585},"will-more-agents-stop-hallucinations","Will more agents stop hallucinations?",[174,588,589],{},"They can catch some errors the way a second reader can. They do not replace asserted wiki, citations, or a person on commitments. Air Canada-style fiction is a gate problem, not a cast-size problem.",[197,591,593],{"id":592},"related-reading","Related reading",[174,595,596,599,600,246],{},[189,597,598],{"href":244},"What is an AI workstream"," and ",[189,601,492],{"href":491},[197,603,605],{"id":604},"sources","Sources",[202,607,608],{},[205,609,610],{},[189,611,613],{"href":191,"rel":612},[193],"Anthropic, Building effective agents",{"title":155,"searchDepth":156,"depth":156,"links":615},[616,617,622,623,637,638],{"id":199,"depth":156,"text":200},{"id":290,"depth":156,"text":291,"children":618},[619,621],{"id":349,"depth":620,"text":350},3,{"id":388,"depth":620,"text":389},{"id":465,"depth":156,"text":466},{"id":495,"depth":156,"text":496,"children":624},[625,626,627,628,629,630,631,632,633,634,635,636],{"id":499,"depth":620,"text":500},{"id":506,"depth":620,"text":507},{"id":513,"depth":620,"text":514},{"id":523,"depth":620,"text":524},{"id":530,"depth":620,"text":531},{"id":537,"depth":620,"text":538},{"id":544,"depth":620,"text":545},{"id":555,"depth":620,"text":556},{"id":562,"depth":620,"text":563},{"id":569,"depth":620,"text":570},{"id":578,"depth":620,"text":579},{"id":585,"depth":620,"text":586},{"id":592,"depth":156,"text":593},{"id":604,"depth":156,"text":605},"2026-08-17","Multi-agent AI is more than one AI specialist handing work to each other — the way legal already reviews a go-to-market draft — with a shared job, a stop, and a person who must approve before a live system changes.","/blog/what-is-multi-agent-ai",{"title":165,"description":640},"explainer","blog/what-is-multi-agent-ai",[643,646,647,648],"multi-agent","agent-teams","enterprise-ai","GptGcIymv6Z_AlmM_K7GOr3c6cofGsabWQwojdUy7Xc",{"hero":651,"id":653,"title":654,"archived":149,"authors":150,"badge":150,"body":655,"date":150,"department":150,"description":659,"extension":158,"eyebrow":660,"faqHeader":150,"faqs":150,"footerBand":661,"headline":150,"image":150,"industry":150,"jobType":150,"listed":149,"location":150,"navigation":115,"openRoles":150,"pageLayout":150,"path":60,"relatedHeading":667,"seo":668,"series":150,"sitemap":115,"status":150,"stem":669,"subhead":150,"tags":150,"video":150,"whyJoin":150,"workplaceType":150,"__hash__":670},{"filename":652},"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":152,"value":656,"toc":657},[],{"title":155,"searchDepth":156,"depth":156,"links":658},[],"Deep dives into pre-cognitive intelligence, sentient enterprises, and the evolving landscape of AI-driven business transformation.","Latest Research",{"headline":662,"description":663,"primaryLabel":664,"primaryTo":665,"secondaryLabel":666,"secondaryTo":12},"Stay at the frontier.","Subscribe for product updates and new insights.","Subscribe","/newsletter","Explore the platform","More research",{"title":654,"description":659},"blog/index","eK1RCXdDW8nfLSyKRXGB1mJm9FAmhAO6GWXwNKOMNVE",[672,1122],{"id":673,"title":674,"archived":149,"authors":675,"badge":677,"body":678,"date":639,"department":150,"description":1113,"extension":158,"eyebrow":150,"faqHeader":150,"faqs":150,"footerBand":150,"headline":150,"image":150,"industry":150,"jobType":150,"listed":149,"location":150,"navigation":115,"openRoles":150,"pageLayout":150,"path":1114,"relatedHeading":150,"seo":1115,"series":643,"sitemap":115,"status":150,"stem":1116,"subhead":150,"tags":1117,"video":150,"whyJoin":150,"workplaceType":150,"__hash__":1121},"content/blog/what-is-shadow-ai.md","What is Shadow AI",[676],{"name":168,"to":120},{"label":170},{"type":152,"value":679,"toc":1088},[680,683,686,693,701,704,706,761,769,771,774,824,827,831,834,837,849,851,856,867,872,877,882,884,890,896,902,908,914,921,929,931,937,940,946,948,952,958,962,965,969,976,980,983,987,990,994,997,1001,1011,1015,1022,1026,1029,1033,1038,1042,1045,1049,1052,1054,1061,1063],[174,681,682],{},"Shadow AI is employees using personal ChatGPT, Claude, Gemini, or similar tools for work because the official company tool is too slow, too locked down, or missing.",[174,684,685],{},"The work is real. The risk is off the books. Security often hears about it first as an incident.",[174,687,688,689,692],{},"It is the AI-era cousin of ",[208,690,691],{},"shadow IT",": unsanctioned software people adopt because it helps them finish the job. The pattern is older than ChatGPT — personal Dropbox, unsanctioned notebooks, Excel macros that became load-bearing. Generative AI sped it up because the tools are excellent, cheap, and one paste away from a customer list.",[174,694,695,700],{},[189,696,699],{"href":697,"rel":698},"https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls",[193],"IBM’s 2025 Cost of a Data Breach research"," found that 20% of organisations reported security incidents involving shadow AI, and that organisations with high levels of it paid $670,000 more per breach. Sixty-three percent lacked AI governance policies.",[174,702,703],{},"Shame does not fix those numbers. Substitution does. People paste work into personal accounts because the deadline is tonight and the official programme is a waitlist. If the approved tool cannot see Salesforce, they will export a spreadsheet. If the approved tool is ten times slower than paste, shadow wins.",[197,705,200],{"id":199},[202,707,708,718,728,734,740,746,755],{},[205,709,710,713,714,717],{},[208,711,712],{},"Shadow IT."," Unsanctioned systems. Shadow AI is often unsanctioned ",[184,715,716],{},"generation"," on a sanctioned laptop — a personal account, not a new product install. At work, the browser is allowed; the tenant is not yours.",[205,719,720,723,724,727],{},[208,721,722],{},"Sanctioned tool."," The company’s official AI, with company login and a vendor agreement. At work, ChatGPT Enterprise on the company tenant can be sanctioned and still be ",[208,725,726],{},"ungoverned for writes"," if people copy output into CRM.",[205,729,730,733],{},[208,731,732],{},"Data leakage."," Prompts become logs at a vendor you have no processing agreement with. At work, a customer list in a consumer chat is a processing event you cannot inventory.",[205,735,736,739],{},[208,737,738],{},"Acceptable-use policy."," A PDF. Necessary. Not a substitute for a tool people can actually use.",[205,741,742,745],{},[208,743,744],{},"DLP / CASB."," Network or cloud tools that watch paste-out and unsanctioned apps. Useful. They do not quote a CRM change or bind a named signer.",[205,747,748,751,752,246],{},[208,749,750],{},"Personal API key."," A pass-through that looks like engineering hygiene and is often shadow AI with a credit card. See ",[189,753,754],{"href":444},"What is AI token economics",[205,756,757,760],{},[208,758,759],{},"Pressure valve."," A logged sandbox with fake data and no production writes. Not shadow. A way to experiment without a customer list.",[174,762,763,768],{},[189,764,767],{"href":765,"rel":766},"https://www.enisa.europa.eu/publications/enisa-threat-landscape-2025",[193],"ENISA’s Threat Landscape 2025"," notes fake AI-tool sites and malware posing as AI installers. People hunting for “a free assistant” are the audience. Blocking the official vendors without a substitute trains that hunt.",[197,770,291],{"id":290},[174,772,773],{},"That can mean:",[202,775,776,788,798,808,818],{},[205,777,778,781,782,787],{},[208,779,780],{},"Customer or internal data sitting in a consumer vendor’s logs."," Legal later asks which model saw it. Nobody can say. ",[189,783,786],{"href":784,"rel":785},"https://eur-lex.europa.eu/eli/reg/2016/679/oj",[193],"GDPR"," does not pause because the employee used a personal account.",[205,789,790,793,794,797],{},[208,791,792],{},"Changes with no record."," Someone types model output into CRM. No approver of record. That is ungoverned ",[189,795,796],{"href":551},"write-back"," with extra steps.",[205,799,800,803,804,246],{},[208,801,802],{},"Two versions of policy."," The official playbook says one thing. A shadow chat invented another. See ",[189,805,807],{"href":806},"what-is-a-company-wiki-for-ai-agents","What is a company wiki for AI agents",[205,809,810,813,814,246],{},[208,811,812],{},"Institutional amnesia."," The reasoning lived in a thread the company cannot query. See ",[189,815,817],{"href":816},"what-is-institutional-memory-in-enterprise-ai","What is institutional memory in enterprise AI",[205,819,820,823],{},[208,821,822],{},"A wider attack surface."," Fake installer sites, prompt leakage, and keys in plugins.",[174,825,826],{},"Blocking websites without offering a sanctioned path does not end shadow AI. It trains people to use personal phones.",[347,828,830],{"id":829},"what-actually-reduces-it","What actually reduces it",[174,832,833],{},"Find the jobs people are already doing in personal chats: drafting, summarising, extracting tables, writing the email. Put those jobs on an official path that can see the right files without a paste.",[174,835,836],{},"Read-only links to live systems in an approved product are how you stop the spreadsheet. Make the official path not much slower than paste. The honest metric is time to finish the job.",[174,838,839,840,843,844,848],{},"Perimeter blocking can tighten ",[184,841,842],{},"after"," a real path exists — not before. ",[189,845,847],{"href":846},"what-is-ai-governance","AI governance"," that is only a block list is guidance with extra steps.",[347,850,350],{"id":349},[174,852,853,855],{},[208,854,355],{}," Shadow spend hides on personal cards and departmental tools. Shadow output typed into the ledger has no trail. Finance should want a sanctioned path with quotes and caps, not a ban that moves the bill onto expenses.",[174,857,858,860,861,866],{},[208,859,361],{}," Processing without an agreement, invented customer commitments, and no inventory of what left the tenant. Air Canada’s chatbot was official and still made a false commitment — ",[189,862,865],{"href":863,"rel":864},"https://www.cbc.ca/news/canada/british-columbia/air-canada-chatbot-lawsuit-1.7116416",[193],"CBC",". Shadow tools add the same class of fiction with even less control. Legal should not allow “non-sensitive only” personal accounts; employees are bad at classifying.",[174,868,869,871],{},[208,870,372],{}," Deadline pressure is the demand signal. Ops should treat missing connectors and waitlists as root causes, and should offer sandboxes so experimentation does not need production data.",[174,873,874,876],{},[208,875,378],{}," Fastest to shadow, because the consumer tools are excellent at email and decks. GTM needs read-only CRM in the official path or they will export. They also copy invented pricing into the opportunity — a write-back problem dressed as productivity.",[174,878,879,881],{},[208,880,384],{}," Detection (surveys, DLP, key scanning) plus substitution. Punishment first yields dishonest surveys. IBM’s uplift in breach cost is the board-level argument; ENISA’s fake-tool landscape is the practical one. A secure web gateway is not a named signer.",[347,883,389],{"id":388},[174,885,886,889],{},[208,887,888],{},"Blocking as strategy."," Phones exist.",[174,891,892,895],{},[208,893,894],{},"Sanctioned equals governed."," Company ChatGPT can still be copy-paste into Salesforce.",[174,897,898,901],{},[208,899,900],{},"Allowing personal accounts for “non-sensitive” work."," Classification fails under deadline.",[174,903,904,907],{},[208,905,906],{},"Shame."," Drives better hiding, not better behaviour.",[174,909,910,913],{},[208,911,912],{},"Assuming shadow is a people problem."," It is usually a missing-path problem: no connectors, no speed, no permission to try.",[174,915,916,917,920],{},"Good looks like: self-service ",[189,918,919],{"href":244},"workstreams",", wiki, read-only connectors, a named signer on writes, time-to-job close to paste, perimeter controls after substitution, sandboxes with fake data. Failure looks like a blocked URL, a PDF, and a personal Claude project full of customers.",[174,922,923,928],{},[189,924,927],{"href":925,"rel":926},"https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/",[193],"ICO guidance on AI and data protection"," still applies when the employee is the one pasting. Lawful basis and purpose do not wait for an official rollout. That is why substitution is a legal control as well as a security one: the unofficial path is still processing.",[197,930,466],{"id":465},[174,932,933,934,936],{},"Nimbus is built so the legitimate path is the easy path: operators open ",[189,935,919],{"href":244}," themselves, use approved playbooks and read-only connectors, and only write to live systems after a named person signs.",[174,938,939],{},"Nimbus does not “detect shadow AI” the way a network tool that watches cloud apps would. Those perimeter tools still matter. The product bet is gravitational: if governed work is live quickly, shadow has less to do.",[174,941,482,942,599,944,246],{},[189,943,39],{"href":40},[189,945,31],{"href":32},[197,947,496],{"id":495},[347,949,951],{"id":950},"is-using-chatgpt-enterprise-still-shadow-ai","Is using ChatGPT Enterprise still shadow AI?",[174,953,954,955,957],{},"If it is the organisation’s tenant, with company login, a processing agreement, and a defined use policy, it is sanctioned — not shadow. It can still be ",[208,956,726],{}," (people copy output into CRM). Sanctioned is not the same as sufficient.",[347,959,961],{"id":960},"does-blocking-openai-at-the-office-network-solve-it","Does blocking OpenAI at the office network solve it?",[174,963,964],{},"It reduces one channel. It does not stop phones, home networks, or other vendors. Without a substitute, it also reduces productivity.",[347,966,968],{"id":967},"can-we-allow-personal-accounts-for-non-sensitive-work","Can we allow personal accounts for “non-sensitive” work?",[174,970,971,972,975],{},"Employees are bad at classifying. If you allow it, assume leakage of whatever they ",[184,973,974],{},"think"," is non-sensitive.",[347,977,979],{"id":978},"how-do-we-find-existing-shadow-ai","How do we find existing shadow AI?",[174,981,982],{},"Anonymous surveys, credit-card review, data-loss monitoring, scanning for personal API keys, and talking to the teams under deadline pressure. Do not start with punishment if you want honest answers.",[347,984,986],{"id":985},"why-do-people-prefer-the-unofficial-tools","Why do people prefer the unofficial tools?",[174,988,989],{},"Speed, quality, missing connectors in the official tool, and fear of “the AI team.” Treat those as requirements, not as moral failure.",[347,991,993],{"id":992},"is-a-logged-sandbox-shadow-ai","Is a logged sandbox shadow AI?",[174,995,996],{},"No. Fake data, no production writes, company login: that is a pressure valve. Production customer lists in a personal account are not.",[347,998,1000],{"id":999},"how-is-this-different-from-shadow-it","How is this different from shadow IT?",[174,1002,1003,1004,1007,1008,1010],{},"Shadow IT is often an unsanctioned ",[184,1005,1006],{},"system",". Shadow AI is often unsanctioned ",[184,1009,716],{}," on a laptop you issued. Your asset inventory will look clean while the prompts leave.",[347,1012,1014],{"id":1013},"what-does-ibms-research-actually-say-here","What does IBM’s research actually say here?",[174,1016,1017,1018,246],{},"IBM reported shadow-AI incidents, higher average breach cost where shadow AI was high, and a large share of organisations lacking AI governance policies. Use it as evidence that this is a control topic, not a manners topic. Read the ",[189,1019,1021],{"href":697,"rel":1020},[193],"newsroom summary",[347,1023,1025],{"id":1024},"will-a-better-acceptable-use-policy-be-enough","Will a better acceptable-use policy be enough?",[174,1027,1028],{},"Write it. Then put the same rules in a product people can finish the job with. PDFs do not see Salesforce.",[347,1030,1032],{"id":1031},"how-do-writes-sneak-in","How do writes sneak in?",[174,1034,1035,1036,246],{},"The model never calls Salesforce. A human pastes the answer. That is still a change to a live system with no quote and no named signer. See ",[189,1037,552],{"href":551},[347,1039,1041],{"id":1040},"should-we-ban-plugins-and-personal-api-keys","Should we ban plugins and personal API keys?",[174,1043,1044],{},"Treat them as unsanctioned processing until they sit on a company path with a ceiling. Keys in wikis are an unmetered utility and a credential incident.",[347,1046,1048],{"id":1047},"what-is-the-first-sanctioned-path-worth-shipping","What is the first sanctioned path worth shipping?",[174,1050,1051],{},"Read-only connectors on the jobs people already paste — email, extract, summarise — plus a wiki they can cite. Writes come later, fail-closed. Speed matters more than a perfect platform launch.",[197,1053,593],{"id":592},[174,1055,1056,599,1059,246],{},[189,1057,1058],{"href":846},"What is AI governance",[189,1060,552],{"href":551},[197,1062,605],{"id":604},[202,1064,1065,1071,1077,1082],{},[205,1066,1067],{},[189,1068,1070],{"href":697,"rel":1069},[193],"IBM newsroom, Cost of a Data Breach 2025",[205,1072,1073],{},[189,1074,1076],{"href":765,"rel":1075},[193],"ENISA Threat Landscape 2025",[205,1078,1079],{},[189,1080,786],{"href":784,"rel":1081},[193],[205,1083,1084],{},[189,1085,1087],{"href":863,"rel":1086},[193],"CBC, Air Canada chatbot lawsuit",{"title":155,"searchDepth":156,"depth":156,"links":1089},[1090,1091,1096,1097,1111,1112],{"id":199,"depth":156,"text":200},{"id":290,"depth":156,"text":291,"children":1092},[1093,1094,1095],{"id":829,"depth":620,"text":830},{"id":349,"depth":620,"text":350},{"id":388,"depth":620,"text":389},{"id":465,"depth":156,"text":466},{"id":495,"depth":156,"text":496,"children":1098},[1099,1100,1101,1102,1103,1104,1105,1106,1107,1108,1109,1110],{"id":950,"depth":620,"text":951},{"id":960,"depth":620,"text":961},{"id":967,"depth":620,"text":968},{"id":978,"depth":620,"text":979},{"id":985,"depth":620,"text":986},{"id":992,"depth":620,"text":993},{"id":999,"depth":620,"text":1000},{"id":1013,"depth":620,"text":1014},{"id":1024,"depth":620,"text":1025},{"id":1031,"depth":620,"text":1032},{"id":1040,"depth":620,"text":1041},{"id":1047,"depth":620,"text":1048},{"id":592,"depth":156,"text":593},{"id":604,"depth":156,"text":605},"Shadow AI is people using personal ChatGPT or similar for work because the official tool is too slow or missing — which leaks data and leaves no record of what changed.","/blog/what-is-shadow-ai",{"title":674,"description":1113},"blog/what-is-shadow-ai",[643,1118,1119,1120],"shadow-ai","governance","security","KN9Y86Ep7MHgpyV42_OuagUFVF0e4uq-AzZGLvkQLX4",{"id":1123,"title":1124,"archived":149,"authors":1125,"badge":1127,"body":1128,"date":639,"department":150,"description":1530,"extension":158,"eyebrow":150,"faqHeader":150,"faqs":150,"footerBand":150,"headline":150,"image":150,"industry":150,"jobType":150,"listed":149,"location":150,"navigation":115,"openRoles":150,"pageLayout":150,"path":1531,"relatedHeading":150,"seo":1532,"series":643,"sitemap":115,"status":150,"stem":1533,"subhead":150,"tags":1534,"video":150,"whyJoin":150,"workplaceType":150,"__hash__":1538},"content/blog/what-is-model-routing.md","What is Model Routing",[1126],{"name":168,"to":120},{"label":170},{"type":152,"value":1129,"toc":1506},[1130,1136,1139,1142,1155,1157,1217,1224,1229,1231,1234,1241,1263,1274,1276,1281,1289,1294,1299,1304,1306,1312,1322,1328,1334,1340,1346,1353,1368,1374,1376,1379,1382,1392,1394,1398,1401,1405,1408,1412,1415,1419,1422,1426,1429,1433,1438,1442,1445,1449,1452,1456,1459,1463,1468,1472,1475,1479,1482,1484,1490,1492],[174,1131,1132,1133,1135],{},"Labs ship a ladder of models: small and cheap, large and expensive. ",[208,1134,262],{}," is the policy above that ladder: use a cheaper, faster model for simple steps, and a stronger model only when the task needs it.",[174,1137,1138],{},"It is not a dropdown labelled “best.” Someone typing “use the best model” for a classify-this-ticket step is how a flagship invoice gets burned on work a compact model could have finished in a second.",[174,1140,1141],{},"Done well, extract runs on compact models and hard reasoning runs on frontier models. Done poorly, every step hits the most expensive model, spend becomes a surprise, and “we use the best model” becomes an unexamined religion.",[174,1143,1144,599,1149,1154],{},[189,1145,1148],{"href":1146,"rel":1147},"https://openai.com/api/pricing/",[193],"OpenAI",[189,1150,1153],{"href":1151,"rel":1152},"https://www.anthropic.com/pricing",[193],"Anthropic"," publish those ladders in public. The prices change. The shape does not: input, output, and sometimes tools meter differently, and the top rung is many times the compact rung. Finance cannot treat “always flagship” as a quality culture. It is an unbudgeted preference.",[197,1156,200],{"id":199},[202,1158,1159,1165,1171,1177,1183,1193,1203,1209],{},[205,1160,1161,1164],{},[208,1162,1163],{},"Frontier / flagship model."," The strongest (and usually most expensive) model a lab currently sells. At work, this is for judgment: does this clause violate the playbook?",[205,1166,1167,1170],{},[208,1168,1169],{},"Compact / small model."," Faster and cheaper. Often enough for extract, classify, and summarise. At work, this is “pull the fields from the export.”",[205,1172,1173,1176],{},[208,1174,1175],{},"Cascade."," Try cheap first; spend the expensive call only when the cheap one is not enough.",[205,1178,1179,1182],{},[208,1180,1181],{},"Fallback."," If a provider is down or over budget, send the step somewhere else.",[205,1184,1185,1188,1189,1192],{},[208,1186,1187],{},"Orchestration."," What steps exist. Different from routing, which is ",[184,1190,1191],{},"which brain"," each step uses. You can orchestrate a brilliant multi-agent graph and still send every node to the flagship.",[205,1194,1195,1198,1199,1202],{},[208,1196,1197],{},"Quality bar."," The reject-rate or rework threshold that decides whether a compact model is good enough on ",[184,1200,1201],{},"this"," job.",[205,1204,1205,1208],{},[208,1206,1207],{},"Data residency / data class."," A cheap endpoint may be forbidden for a class of records. Routing is then a compliance table, not only a cost table.",[205,1210,1211,1214,1215,246],{},[208,1212,1213],{},"NTU."," The normalised unit routing is trying to protect. See ",[189,1216,754],{"href":444},[174,1218,1219,1220,1223],{},"Routing is also not ",[208,1221,1222],{},"fine-tuning"," (changing a model’s weights). Fine-tuning is a research and ops programme. Routing is an operating policy over models you already buy.",[174,1225,1226,1227,246],{},"Constraints that belong in the route table: data residency, evaluation (you cannot route on vibes), and security (a model with web tools is a different actor than a model with none). Choosing a model is choosing a brain. Choosing tools is choosing hands. Decide them separately. See ",[189,1228,278],{"href":277},[197,1230,291],{"id":290},[174,1232,1233],{},"It affects you if you pay the bill, or if quality on a step is load-bearing.",[174,1235,1236,1237,1240],{},"Talk about it as a ",[208,1238,1239],{},"budget and quality conversation",", not as an ML research project:",[202,1242,1243,1249,1255],{},[205,1244,1245,1248],{},[208,1246,1247],{},"Tag the steps."," Extracting fields from an export is not the same as arguing whether a clause violates policy. If your platform cannot name steps, it cannot route them.",[205,1250,1251,1254],{},[208,1252,1253],{},"Set a quality bar per step."," “Compact model until human reject rate exceeds X on this job.” Without a bar, routing becomes “always escalate because someone was once unhappy.”",[205,1256,1257,1260,1261,246],{},[208,1258,1259],{},"Keep the gate regardless of model."," A cheap model with a write tool is still a write tool. See ",[189,1262,552],{"href":551},[174,1264,1265,1266,1269,1270,1273],{},"A spend ceiling without routing still lets every step hit the flagship until the ceiling kills the run. Routing is how you stay under the ceiling ",[184,1267,1268],{},"and"," finish the job. See ",[189,1271,1272],{"href":285},"What is an agentic workflow"," for why loops without stops dominate the bill.",[347,1275,350],{"id":349},[174,1277,1278,1280],{},[208,1279,355],{}," Routing is the practical lever on unit cost. Quotes should assume the policy, not the flagship. Finance should ask for approved-updates per NTU, and for evidence that extract steps are not on the top rung. Locking one vendor forever is a pricing and outage choice; routing across providers is a second tape measure.",[174,1282,1283,1285,1286,1288],{},[208,1284,361],{}," Data class and residency can forbid the cheap endpoint. Legal should sit on the route table for those classes, not discover them on an invoice. Customer-facing language may need a stronger model ",[184,1287,1268],{}," a named signer; routing does not replace the gate.",[174,1290,1291,1293],{},[208,1292,372],{}," Steps must be named or you cannot route them. Ops should own fallbacks when a provider is down, and should refuse a single “best” toggle that bypasses the table.",[174,1295,1296,1298],{},[208,1297,378],{}," Quality anxiety is strongest here. Measure reject rates on the job. A compact model that extracts next steps may be fine; a compact model that invents a concession is not. Routing on one unhappy anecdote will pin every step to flagship.",[174,1300,1301,1303],{},[208,1302,384],{}," A model with browsing or unconstrained tools is a different actor. Routing should not silently add hands. Prompt injection plus a flagship model plus write tools is a worse combination than a compact extract-only step behind a fail-closed gate.",[347,1305,389],{"id":388},[174,1307,1308,1311],{},[208,1309,1310],{},"Always the smartest model."," Use the weakest model that meets the quality bar for that step. Flagship is for judgment, not for labelling.",[174,1313,1314,1317,1318,246],{},[208,1315,1316],{},"Routing as multi-agent."," Several agents is a cast. Routing is which brain each step pays for. See ",[189,1319,1321],{"href":1320},"what-is-multi-agent-ai","What is multi-agent AI",[174,1323,1324,1327],{},[208,1325,1326],{},"Routing as fine-tuning."," Different programme.",[174,1329,1330,1333],{},[208,1331,1332],{},"Dropdown labelled “best.”"," That is not a policy. It is a preference that cannot be audited.",[174,1335,1336,1339],{},[208,1337,1338],{},"Dropping the write gate for a “trusted” model."," Trust the gate. Models change weekly.",[174,1341,1342,1345],{},[208,1343,1344],{},"Routing on vibes."," One anecdote becomes a permanent escalate. Measure rework.",[174,1347,1348,1349,1352],{},"Good looks like: named steps, a route table with cost, quality bar, and data class, cascade where it helps, fallback across providers, gates independent of model, falling unit cost as the ",[189,1350,1351],{"href":806},"wiki"," reduces re-derivation. Failure looks like flagship-everywhere and a board slide about the bill.",[174,1354,1355,1356,1359,1360,1363,1364,1367],{},"Adjacent ideas worth keeping separate: ",[189,1357,1358],{"href":444},"token economics"," is quote, cap, and attribute. Routing is which rung of the ladder a named step is allowed to use. ",[189,1361,1362],{"href":1320},"Multi-agent AI"," is how many specialist roles run. You can route a single agent, and you can send a whole agent team to the flagship by mistake. ",[189,1365,1366],{"href":277},"MCP"," is hands, not brains: do not let a compact extract step inherit a write tool because “the helper was available.”",[174,1369,1370,1371,1373],{},"Evaluation has to live on the job, not in a model-arena screenshot. A compact model that extracts fields with a low reject rate is a success even if it would lose a public chatbot bake-off. A flagship model that drafts a concession the wiki forbids is a failure even if it is eloquent. Tie routing reviews to ",[189,1372,245],{"href":244}," outcomes — approved writes, rejects, rework — the same way you would review any other operating policy.",[197,1375,466],{"id":465},[174,1377,1378],{},"Nimbus treats routing as an operating decision tied to workstream steps: task type, sensitivity, and cost — not “best everywhere.” Release gates apply regardless of which model drafted the payload.",[174,1380,1381],{},"NTU quotes and ceilings sit around that policy so operators see a number before they commit. Everyday extract should not consume flagship credits.",[174,1383,482,1384,1386,1387,1389,1390,246],{},[189,1385,44],{"href":45},". For the unit of account routing sits inside, ",[189,1388,754],{"href":444},". Product: ",[189,1391,31],{"href":32},[197,1393,496],{"id":495},[347,1395,1397],{"id":1396},"should-we-always-use-the-smartest-model","Should we always use the smartest model?",[174,1399,1400],{},"No. Use the weakest model that meets the quality bar for that step. Flagship is for judgment, not for labelling.",[347,1402,1404],{"id":1403},"will-routing-make-answers-worse","Will routing make answers worse?",[174,1406,1407],{},"It can, if you under-route hard steps. Measure rejects and rework on the job. Do not route on a single anecdote.",[347,1409,1411],{"id":1410},"is-this-the-same-as-having-several-agents","Is this the same as having several agents?",[174,1413,1414],{},"No. Several agents is a cast. Routing is which brain each step pays for.",[347,1416,1418],{"id":1417},"can-we-lock-one-vendor-forever","Can we lock one vendor forever?",[174,1420,1421],{},"You can. You will pay for it in price, outages, and lock-in. Routing across providers is how finance keeps a second tape measure.",[347,1423,1425],{"id":1424},"what-is-a-cascade","What is a cascade?",[174,1427,1428],{},"Try the cheap model first. Escalate only when a confidence or quality check says the cheap pass is not enough. It is a tactic inside a policy, not a substitute for naming steps.",[347,1430,1432],{"id":1431},"does-a-better-model-remove-the-need-for-a-wiki","Does a better model remove the need for a wiki?",[174,1434,1435,1436,246],{},"No. Stronger models are better at sounding like policy. Asserted playbooks still win over Drive folklore. See ",[189,1437,807],{"href":806},[347,1439,1441],{"id":1440},"how-do-we-set-a-quality-bar","How do we set a quality bar?",[174,1443,1444],{},"Start with human reject rate and rework on that step. “Compact until rejects exceed X on this workstream” is a bar. “People like the flagship” is not.",[347,1446,1448],{"id":1447},"should-customer-facing-copy-always-use-the-flagship","Should customer-facing copy always use the flagship?",[174,1450,1451],{},"Not always. It should always use a human gate if it asserts a term or a price. Model size does not absorb Air Canada-style risk.",[347,1453,1455],{"id":1454},"what-if-the-cheap-endpoint-is-in-the-wrong-region","What if the cheap endpoint is in the wrong region?",[174,1457,1458],{},"Then it is not cheap; it is forbidden. Put residency in the route table beside price.",[347,1460,1462],{"id":1461},"does-routing-replace-spend-caps","Does routing replace spend caps?",[174,1464,1465,1466,246],{},"No. Caps stop unbounded loops. Routing makes legitimate work affordable under the cap. You want both. See ",[189,1467,754],{"href":444},[347,1469,1471],{"id":1470},"can-the-model-choose-its-own-successor","Can the model choose its own successor?",[174,1473,1474],{},"Letting the model always escalate is how every step becomes flagship. Escalation should be a policy check, not a preference the model expresses.",[347,1476,1478],{"id":1477},"how-does-this-show-up-in-an-rfp","How does this show up in an RFP?",[174,1480,1481],{},"Ask whether steps are named, whether gates apply regardless of model, and whether finance sees a normalised unit. “We use the best models” is not an answer.",[197,1483,593],{"id":592},[174,1485,1486,599,1488,246],{},[189,1487,754],{"href":444},[189,1489,1321],{"href":1320},[197,1491,605],{"id":604},[202,1493,1494,1500],{},[205,1495,1496],{},[189,1497,1499],{"href":1146,"rel":1498},[193],"OpenAI API pricing",[205,1501,1502],{},[189,1503,1505],{"href":1151,"rel":1504},[193],"Anthropic pricing",{"title":155,"searchDepth":156,"depth":156,"links":1507},[1508,1509,1513,1514,1528,1529],{"id":199,"depth":156,"text":200},{"id":290,"depth":156,"text":291,"children":1510},[1511,1512],{"id":349,"depth":620,"text":350},{"id":388,"depth":620,"text":389},{"id":465,"depth":156,"text":466},{"id":495,"depth":156,"text":496,"children":1515},[1516,1517,1518,1519,1520,1521,1522,1523,1524,1525,1526,1527],{"id":1396,"depth":620,"text":1397},{"id":1403,"depth":620,"text":1404},{"id":1410,"depth":620,"text":1411},{"id":1417,"depth":620,"text":1418},{"id":1424,"depth":620,"text":1425},{"id":1431,"depth":620,"text":1432},{"id":1440,"depth":620,"text":1441},{"id":1447,"depth":620,"text":1448},{"id":1454,"depth":620,"text":1455},{"id":1461,"depth":620,"text":1462},{"id":1470,"depth":620,"text":1471},{"id":1477,"depth":620,"text":1478},{"id":592,"depth":156,"text":593},{"id":604,"depth":156,"text":605},"Model routing is using a cheaper, faster model for simple steps and a stronger model only when the task needs it — a policy, not a dropdown labelled “best.”","/blog/what-is-model-routing",{"title":1124,"description":1530},"blog/what-is-model-routing",[643,1535,1536,1537],"model-routing","token-economics","cost","cT8uO7Fl6bffuYVI7nA4O6i99dYBnBc0RAb25dGjpU0",{"enabled":149,"message":1540,"linkLabel":78,"linkHref":79,"id":1541,"title":1542,"archived":149,"authors":150,"badge":150,"body":1543,"date":150,"department":150,"description":155,"extension":158,"eyebrow":150,"faqHeader":150,"faqs":150,"footerBand":150,"headline":150,"image":150,"industry":150,"jobType":150,"listed":149,"location":150,"navigation":115,"openRoles":150,"pageLayout":150,"path":1547,"relatedHeading":150,"seo":1548,"series":150,"sitemap":115,"status":150,"stem":1549,"subhead":150,"tags":150,"video":150,"whyJoin":150,"workplaceType":150,"__hash__":1550},"We're hiring! Join the team building the Sentient Enterprise.","content/shared/hiring.md","Hiring banner",{"type":152,"value":1544,"toc":1545},[],{"title":155,"searchDepth":156,"depth":156,"links":1546},[],"/shared/hiring",{"title":1542,"description":155},"shared/hiring","-6bioYD7lKYokGUVU3ff4hHTvB-sDyOMuCptKHnojfk",{"fold":1552,"id":1556,"title":1557,"archived":149,"authors":150,"badge":150,"body":1558,"date":150,"department":150,"description":155,"extension":158,"eyebrow":150,"faqHeader":150,"faqs":150,"footerBand":1562,"headline":150,"image":150,"industry":150,"jobType":150,"listed":149,"location":150,"navigation":115,"openRoles":150,"pageLayout":150,"path":1566,"relatedHeading":150,"seo":1567,"series":150,"sitemap":115,"status":150,"stem":1568,"subhead":150,"tags":150,"video":150,"whyJoin":150,"workplaceType":150,"__hash__":1569},{"headline":1553,"description":1554,"primaryLabel":8,"primaryTo":1555,"secondaryLabel":666,"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":152,"value":1559,"toc":1560},[],{"title":155,"searchDepth":156,"depth":156,"links":1561},[],{"headline":1563,"description":1564,"primaryLabel":8,"primaryTo":1555,"secondaryLabel":1565,"secondaryTo":84},"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":1557,"description":155},"shared/cta","YHK6Fb8AvCPR1zZq7R_xiXUG0hwhP5UxHA8Ix52JQp4",1787194074849]