[{"data":1,"prerenderedAt":3496},["ShallowReactive",2],{"site-nav-content":3,"blog:/blog/what-nimbus-replaces-in-your-stack":179,"blog-index-copy":1468,"blog:/blog/what-nimbus-replaces-in-your-stack:surround":1489,"hiring-banner-content":3465,"site-cta-content":3477},{"header":4,"productNav":9,"nav":42,"footer":61,"askAI":132,"id":163,"title":164,"archived":165,"authors":166,"badge":166,"body":167,"date":166,"definedTerm":166,"department":166,"description":171,"extension":174,"eyebrow":166,"faqHeader":166,"faqs":166,"footerBand":166,"headline":166,"image":166,"industry":166,"jobType":166,"listed":131,"location":166,"navigation":131,"openRoles":166,"pageLayout":166,"path":175,"relatedHeading":166,"seo":176,"series":166,"sitemap":165,"status":166,"stem":177,"subhead":166,"tags":166,"video":166,"whyJoin":166,"workplaceType":166,"__hash__":178},{"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,"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://docs.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","/careers",{"label":82,"to":83},"System status","/status",{"label":7,"to":85},"/contact",[87,90,93,96,99],{"label":88,"to":89},"Glossary","/glossary",{"label":91,"to":92},"Compare","/compare",{"label":94,"to":95},"Evaluate","/evaluate",{"label":97,"to":98},"Problems","/problems",{"label":100,"to":101},"Use cases","/use-cases",[103,106],{"label":104,"to":105},"Terms of Service","/terms",{"label":107,"to":108},"Privacy Policy","/privacy",[110,113,116],{"label":111,"href":112},"LinkedIn","https://www.linkedin.com/company/gonimbusai/",{"label":114,"href":115},"X","https://x.com/gonimbusai",{"label":117,"href":118},"Instagram","https://www.instagram.com/gonimbus_ai/",[120,122,124,127,128],{"label":121,"to":105},"Terms",{"label":123,"to":108},"Privacy",{"label":125,"to":126},"Compliance","/compliance",{"label":82,"to":83},{"label":129,"to":130,"external":131},"LLMs.txt","/llms.txt",true,{"text":133,"prompt":134},"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":135,"platforms":137},{" Summarize the highlights from Nimbus's website":136},"https://gonimbus.ai",[138,143,148,153,158],{"name":139,"label":140,"icon":141,"hrefPrefix":142},"chatgpt","ChatGPT","simple-icons:openai","https://chatgpt.com/?prompt=",{"name":144,"label":145,"icon":146,"hrefPrefix":147},"perplexity","Perplexity","mdi:magnify","https://www.perplexity.ai/search/new?q=",{"name":149,"label":150,"icon":151,"hrefPrefix":152},"grok","Grok","simple-icons:x","https://x.com/i/grok?text=",{"name":154,"label":155,"icon":156,"hrefPrefix":157},"claude","Claude","simple-icons:anthropic","https://claude.ai/new?q=",{"name":159,"label":160,"icon":161,"hrefPrefix":162},"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":168,"value":169,"toc":170},"minimark",[],{"title":171,"searchDepth":172,"depth":172,"links":173},"",2,[],"md","/shared/nav",{"title":164,"description":171},"shared/nav","1dD7ahDRl0SQ4hz53-kKo0tEFrGaLuaztZ3PPfp6a9k",{"id":180,"title":181,"archived":165,"authors":182,"badge":185,"body":187,"date":1439,"definedTerm":166,"department":166,"description":1440,"extension":174,"eyebrow":166,"faqHeader":1441,"faqs":1444,"footerBand":166,"headline":166,"image":166,"industry":166,"jobType":166,"listed":131,"location":166,"navigation":131,"openRoles":166,"pageLayout":166,"path":1457,"relatedHeading":166,"seo":1458,"series":1459,"sitemap":131,"status":166,"stem":1460,"subhead":166,"tags":1461,"video":166,"whyJoin":166,"workplaceType":166,"__hash__":1467},"content/blog/what-nimbus-replaces-in-your-stack.md","What Nimbus replaces in your current stack",[183],{"name":184,"to":136},"Nimbus Research",{"label":186},"Comparisons",{"type":168,"value":188,"toc":1423},[189,193,207,217,222,225,339,362,366,369,386,394,418,427,431,438,453,528,535,546,550,568,580,668,681,685,688,751,762,765,769,778,788,802,818,831,841,868,872,897,907,1007,1026,1029,1034,1127,1134,1138,1141,1243,1260,1267,1271,1274,1312,1319,1323,1334,1358,1369,1388,1392],[190,191,192],"p",{},"Most companies did not choose an AI stack. They accumulated one. Slack or Teams became the filing cabinet. A search product became “the knowledge layer.” Microsoft 365 Copilot became the programme. ChatGPT or Claude became the unofficial thinking surface. A studio or a Power Platform flow became “agents.” None of those products is a bad tool. Each was stretched into a job it was not designed to finish.",[190,194,195,196,201,202,206],{},"Nimbus is an ",[197,198,200],"a",{"href":199},"what-is-an-enterprise-ai-operating-system","enterprise AI operating system",": isolation of jobs, scoped tools, fail-closed writes, a budget, and a record that survives the session. It does not replace your CRM or try to be Word. It replaces the ",[203,204,205],"em",{},"overlay"," — the products you bought so AI work would have somewhere to live, which still leave the signed change in someone’s private thread.",[190,208,209,210,216],{},"McKinsey’s ",[197,211,215],{"href":212,"rel":213},"https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai",[214],"nofollow","2025 State of AI"," keeps showing the same split: use is common, scale is not. Scale fails when find-and-draft is licensed per head and execution is still folklore. This page maps what to stop stretching, what to keep, and why one operating layer is usually cheaper than the overlay — and better at the jobs the overlay never owned.",[218,219,221],"h2",{"id":220},"the-stack-you-already-paid-for","The stack you already paid for",[190,223,224],{},"A typical “we have AI” environment looks like this:",[226,227,228,247],"table",{},[229,230,231],"thead",{},[232,233,234,238,241,244],"tr",{},[235,236,237],"th",{},"Layer you bought",[235,239,240],{},"Product you actually use",[235,242,243],{},"Job it is good at",[235,245,246],{},"Job it gets stretched into",[248,249,250,265,283,300,321],"tbody",{},[232,251,252,256,259,262],{},[253,254,255],"td",{},"Chat",[253,257,258],{},"Slack or Microsoft Teams",[253,260,261],{},"Conversation",[253,263,264],{},"System of work, memory, approvals",[232,266,267,270,277,280],{},[253,268,269],{},"Context / search",[253,271,272,276],{},[197,273,275],{"href":274},"nimbus-vs-glean","Glean",", SharePoint, Notion",[253,278,279],{},"Find the file",[253,281,282],{},"Operating layer, agents, writes",[232,284,285,288,294,297],{},[253,286,287],{},"Personal copilot",[253,289,290],{},[197,291,293],{"href":292},"nimbus-vs-microsoft-copilot","Microsoft 365 Copilot",[253,295,296],{},"Recap and draft in Office",[253,298,299],{},"Company AI programme",[232,301,302,305,315,318],{},[253,303,304],{},"Hosted assistant",[253,306,307,311,312],{},[197,308,310],{"href":309},"nimbus-vs-chatgpt-enterprise","ChatGPT Enterprise",", ",[197,313,155],{"href":314},"nimbus-vs-claude",[253,316,317],{},"Think with company login",[253,319,320],{},"Place CRM fields change",[232,322,323,326,333,336],{},[253,324,325],{},"Agent studio",[253,327,328,332],{},[197,329,331],{"href":330},"nimbus-vs-dust","Dust",", Copilot Studio",[253,334,335],{},"Publish a helper",[253,337,338],{},"Cross-department job with a signer",[190,340,341,342,346,347,351,352,356,357,361],{},"None of the left-hand products disappears because Nimbus exists. What disappears is the pretence that the right-hand column is free. ",[197,343,345],{"href":344},"search-is-not-memory","Search is not memory",". A recap is not a ",[197,348,350],{"href":349},"what-is-write-back-governance","write-back",". A channel is not a ",[197,353,355],{"href":354},"what-is-an-ai-workstream","workstream",". The ",[197,358,360],{"href":359},"four-pillars-of-an-enterprise-ai-platform","four pillars"," — Communication, Collaboration, Automate, Governance — are the jobs the overlay split across vendors. Nimbus hosts all four on one job object.",[218,363,365],{"id":364},"what-nimbus-replaces-in-slack-and-microsoft-teams","What Nimbus replaces in Slack and Microsoft Teams",[190,367,368],{},"Slack and Teams are excellent at conversation. They are a poor system of record for decisions, and a worse runtime for agents.",[190,370,371,372,376,377,380,381,385],{},"What actually happens: the discount exception lives in a DM. The hold reason lives in ",[373,374,375],"code",{},"#warehouse",". “Legal is comfortable” is a thumbs-up from whoever was online. Next quarter, finance searches the characters ",[203,378,379],{},"p-r-i-c-e"," and calls the hit memory. ",[197,382,384],{"href":383},"decisions-made-in-direct-messages","Decisions made in direct messages"," is that cost. It is not a chat outage. It is a missing object.",[190,387,388,389,393],{},"Nimbus replaces Slack and Teams ",[390,391,392],"strong",{},"as the place AI work lives",":",[395,396,397,404,407,415],"ul",{},[398,399,400,401,403],"li",{},"A named ",[197,402,355],{"href":32},", not a channel with a bot",[398,405,406],{},"Files bound to the job; guests without every connector",[398,408,409,410,414],{},"A ",[197,411,413],{"href":412},"what-is-a-nimbus-loop","loop"," that notifies a roster, not a dead channel",[398,416,417],{},"A refused write as a record, not “Sarah said it looked fine”",[190,419,420,422,423,426],{},[197,421,15],{"href":16}," is the workspace hub — orientation, then a handoff into governed execution. Keep Slack for “are you free at 3.” Stop using it as the ledger. ",[390,424,425],{},"Operator test:"," hide the channel. Can a stranger reconstruct the signer, the payload, and the policy version? If the answer is Slack search, you have folklore with push notifications.",[218,428,430],{"id":429},"what-nimbus-replaces-in-glean-and-other-context-layers","What Nimbus replaces in Glean and other context layers",[190,432,433,437],{},[197,434,275],{"href":435,"rel":436},"https://www.glean.com/",[214]," is permission-aware workplace search. SharePoint search, Notion AI, and “company knowledge” in ChatGPT are cousins: indexes that try not to leak. Nimbus does not try to crawl 2019.",[190,439,440,441,444,445,448,449,452],{},"The stretch is the problem. Search vendors added assistants, then agents, then production credentials “because the index already respects permissions.” Permission to ",[203,442,443],{},"see"," is not permission to ",[203,446,447],{},"change",". ",[197,450,451],{"href":274},"Nimbus vs Glean"," is the full cut. The short version:",[226,454,455,468],{},[229,456,457],{},[232,458,459,462,465],{},[235,460,461],{},"Question",[235,463,464],{},"Context layer (Glean-class)",[235,466,467],{},"Nimbus",[248,469,470,481,492,506,517],{},[232,471,472,475,478],{},[253,473,474],{},"Primary object",[253,476,477],{},"Document, message, person",[253,479,480],{},"Job, team, release",[232,482,483,486,489],{},[253,484,485],{},"Success metric",[253,487,488],{},"Time-to-answer",[253,490,491],{},"Time-to-signed-off outcome",[232,493,494,497,500],{},[253,495,496],{},"Graph",[253,498,499],{},"People and files for retrieval",[253,501,502,505],{},[197,503,23],{"href":504},"what-is-a-lifecycle-graph"," of work and writes",[232,507,508,511,514],{},[253,509,510],{},"Write path",[253,512,513],{},"Agents on the index",[253,515,516],{},"Quoted release, named signer",[232,518,519,522,525],{},[253,520,521],{},"Deployment",[253,523,524],{},"Crawl and permission QA",[253,526,527],{},"Self-serve workspace, wiki, connectors",[190,529,530,531,534],{},"Nimbus replaces the context layer ",[390,532,533],{},"as an operating system",". It does not replace Glean as a universal search bar across a 20,000-person corpus. Keep that knowledge programme if you have one. If RevOps needed a signed opportunity update this quarter and was told to wait for the crawl, you bought the wrong graph.",[190,536,537,540,541,545],{},[197,538,539],{"href":36},"Perception"," is ordinary language over the scoped world you attached. The ",[197,542,544],{"href":543},"what-is-a-company-wiki-for-ai-agents","wiki"," is what the company asserts. Search cannot invent a signer.",[218,547,549],{"id":548},"what-nimbus-replaces-in-microsoft-copilot","What Nimbus replaces in Microsoft Copilot",[190,551,552,553,559,560,567],{},"“Copilot” is several products. ",[390,554,555],{},[197,556,293],{"href":557,"rel":558},"https://www.microsoft.com/microsoft-365/copilot",[214]," is the $30/user/month add-on in Word, Outlook, Teams, and the Copilot app, grounded in Microsoft Graph. ",[390,561,562],{},[197,563,566],{"href":564,"rel":565},"https://learn.microsoft.com/en-us/microsoft-copilot-studio/",[214],"Copilot Studio"," is the low-code environment for agents that leave Microsoft. Collapsing those in procurement is how E5 gets treated as an operating layer.",[190,569,570,571,574,575,579],{},"Nobody else will recap this thread and the attached deck as well as Copilot inside Outlook. That recap is tenant productivity. Updating forty opportunities in Salesforce is company operations. ",[197,572,573],{"href":292},"Nimbus vs Microsoft Copilot"," is the product cut. This page is the stack cut: Copilot is the personal assistant. Nimbus is ",[197,576,578],{"href":577},"what-is-collaborative-ai","collaborative AI"," — several people, one job, a named stop.",[226,581,582,594],{},[229,583,584],{},[232,585,586,588,590,592],{},[235,587,461],{},[235,589,293],{},[235,591,566],{},[235,593,467],{},[248,595,596,610,624,637,655],{},[232,597,598,601,604,607],{},[253,599,600],{},"Home",[253,602,603],{},"Word, Outlook, Teams",[253,605,606],{},"Power Platform",[253,608,609],{},"Workstream",[232,611,612,615,618,621],{},[253,613,614],{},"Corpus",[253,616,617],{},"Microsoft Graph",[253,619,620],{},"Graph plus connectors",[253,622,623],{},"Wiki + scoped connectors",[232,625,626,629,632,635],{},[253,627,628],{},"Write in Salesforce",[253,630,631],{},"Not the product",[253,633,634],{},"Flow-shaped side effect",[253,636,516],{},[232,638,639,642,645,648],{},[253,640,641],{},"Metering",[253,643,644],{},"Seat (+ Studio credits)",[253,646,647],{},"Credits / messages",[253,649,650,651],{},"Pooled ",[197,652,654],{"href":653},"what-is-ai-token-economics","NTUs",[232,656,657,660,663,666],{},[253,658,659],{},"Record",[253,661,662],{},"Tenant productivity",[253,664,665],{},"Flow run",[253,667,23],{},[190,669,670,671,675,676,680],{},"Keep Copilot in Office. What Nimbus replaces is the sentence “every knowledge worker has a Copilot licence, so we have an AI operating model.” E5 explains identity, compliance, and Office. It does not explain CRM write-back, ",[197,672,674],{"href":673},"what-is-model-routing","model routing",", or a company work ledger. Stay personal for solo drafting — see ",[197,677,679],{"href":678},"collaborative-ai-and-personal-assistants","collaborative AI and personal assistants",". Cross-department work that mutates a live system is not that job.",[218,682,684],{"id":683},"what-else-the-overlay-usually-includes","What else the overlay usually includes",[190,686,687],{},"The rest of the budget usually went here:",[395,689,690,706,714,727,737,745],{},[398,691,692,700,701,705],{},[390,693,694,696,697,699],{},[197,695,310],{"href":309}," and ",[197,698,155],{"href":314},"."," The right move against ",[197,702,704],{"href":703},"what-is-shadow-ai","shadow AI",". Still an assistant. Confirmation-gated writes are not a release.",[398,707,708,713],{},[390,709,710,712],{},[197,711,331],{"href":330}," and other studios."," Shared helpers. Activity logs are not a signed CRM change.",[398,715,716,722,723,726],{},[390,717,718,699],{},[197,719,721],{"href":720},"nimbus-vs-salesforce-agentforce","Salesforce Agentforce"," The right agent ",[203,724,725],{},"inside"," Salesforce. Jobs that also involve Drive and legal still need a place that is not only field history.",[398,728,729,732,733,699],{},[390,730,731],{},"RPA."," Keep it where the screen is the only door. It is not a loop, a workflow, or a reasoner. See ",[197,734,736],{"href":735},"loop-vs-rpa","loop vs RPA",[398,738,739,744],{},[390,740,741,699],{},[197,742,145],{"href":743},"nimbus-vs-perplexity"," Cited answers from the web. Not a production login.",[398,746,747,750],{},[390,748,749],{},"Zapier / Make."," Fine for a notification. Not a roster, a wiki clause, or a fail-closed write.",[190,752,753,754,758,759,761],{},"The composition you actually want is in ",[197,755,757],{"href":756},"loop-vs-workflow-vs-agent","loop vs workflow vs agent",": agent for the unknown path, workflow when stops and roles matter, loop when the same signal fires again. All three on one ",[197,760,355],{"href":32},". Not three more vendors.",[190,763,764],{},"Drafts can travel. Write credentials should not.",[218,766,768],{"id":767},"why-the-replacement-is-better-not-only-cheaper","Why the replacement is better, not only cheaper",[190,770,771,774,775,777],{},[390,772,773],{},"The job is the unit."," A ",[197,776,355],{"href":354}," has a brief, a roster, scoped connectors, a budget, and a finish line. A channel has none of those. A copilot thread has a user.",[190,779,780,783,784,787],{},[390,781,782],{},"Writes are releases."," Connectors stay read-only until a named person signs a quoted payload. Missing signer means nothing happens. That is ",[197,785,786],{"href":349},"write-back governance",", not a confirmation click people train themselves to ignore.",[190,789,790,793,794,796,797,801],{},[390,791,792],{},"Playbooks are asserted."," The ",[197,795,544],{"href":28}," is what agents must follow — a discount floor, a journal policy — rather than a prompt pasted into a custom GPT. ",[197,798,800],{"href":799},"rbac-for-enterprise-ai","RBAC"," sits on the job.",[190,803,804,807,808,812,813,817],{},[390,805,806],{},"Repeat work compiles."," ",[197,809,811],{"href":810},"what-is-loop-engineering","Loop engineering"," turns Monday’s close into a standing order with a run page. You stop paying a frontier model to rediscover a checklist. ",[197,814,816],{"href":815},"agents-should-be-disposable","Agents should be disposable",": the job survives the model swap.",[190,819,820,793,823,825,826,830],{},[390,821,822],{},"The record is causal.",[197,824,23],{"href":24}," answers “who approved this, against which policy, what changed.” Chat logs answer a different question. ",[197,827,829],{"href":828},"governance-as-a-multiplayer-primitive","Governance as a multiplayer primitive"," is why a control only one person can operate is not an organisational control.",[190,832,833,836,837,840],{},[390,834,835],{},"Context compounds; models are infrastructure."," Last quarter’s signed artefact is input to this quarter’s run. Nimbus ",[197,838,839],{"href":45},"routes"," compact extract versus frontier judgement. You do not re-explain the company to a new chat window every Monday.",[190,842,843,844,311,848,311,852,311,856,311,860,311,864,699],{},"Function walkthroughs: ",[197,845,847],{"href":846},"collaborative-ai-for-operations","operations",[197,849,851],{"href":850},"collaborative-ai-for-revenue-operations","revenue",[197,853,855],{"href":854},"collaborative-ai-for-finance-and-planning","finance",[197,857,859],{"href":858},"collaborative-ai-for-legal-and-compliance-review","legal",[197,861,863],{"href":862},"collaborative-ai-for-human-resources","HR",[197,865,867],{"href":866},"collaborative-ai-for-customer-support","customer support",[218,869,871],{"id":870},"the-pricing-differential-cheaper-because-it-is-a-different-object","The pricing differential — cheaper because it is a different object",[190,873,874,875,880,881,884,885,888,889,892,893,896],{},"Seat-versus-seat comparisons hide the bill. Copilot is ",[197,876,879],{"href":877,"rel":878},"https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/enterprise",[214],"listed"," at ",[390,882,883],{},"$30 per user per month",", paid yearly, on top of a qualifying Microsoft 365 plan. Glean does not publish a rate card; third-party estimates cluster around ",[390,886,887],{},"$45–$50 per user per month",", often with a large-seat minimum. ChatGPT Enterprise is quote-only; reported deals often land near ",[390,890,891],{},"$45–$75 per user per month",". Slack Business+ is ",[390,894,895],{},"$15 per user per month"," annually — and baked-in AI does not make the channel a ledger.",[190,898,899,900,903,904,699],{},"Nimbus is priced per operator, with NTUs that ",[390,901,902],{},"grow with seats and pool across the org",". See ",[197,905,906],{"href":48},"pricing",[226,908,909,925],{},[229,910,911],{},[232,912,913,916,919,922],{},[235,914,915],{},"Plan",[235,917,918],{},"List (annual / monthly)",[235,920,921],{},"Included NTUs",[235,923,924],{},"Effective included rate",[248,926,927,941,961,979,993],{},[232,928,929,932,935,938],{},[253,930,931],{},"Free",[253,933,934],{},"$0",[253,936,937],{},"100 lifetime per user, pooled",[253,939,940],{},"Onboarding, not a usage SKU",[232,942,943,946,952,958],{},[253,944,945],{},"Team",[253,947,948,951],{},[390,949,950],{},"$35 / $44"," per user",[253,953,954,957],{},[390,955,956],{},"500 / user / month",", pooled",[253,959,960],{},"$0.07 / NTU",[232,962,963,966,971,976],{},[253,964,965],{},"Scale",[253,967,968,951],{},[390,969,970],{},"$120 / $150",[253,972,973,957],{},[390,974,975],{},"2,000 / user / month",[253,977,978],{},"$0.06 / NTU",[232,980,981,984,987,990],{},[253,982,983],{},"Enterprise",[253,985,986],{},"Custom",[253,988,989],{},"Committed pool",[253,991,992],{},"Negotiated",[232,994,995,998,1001,1004],{},[253,996,997],{},"Extra work",[253,999,1000],{},"PAYG ~$0.10 / NTU, or a monthly commitment at $0.06",[253,1002,1003],{},"Metered on top of the pool",[253,1005,1006],{},"Published rate, not a surprise overage",[190,1008,1009,1010,1013,1014,1017,1018,1021,1022,1025],{},"A five-seat Team org has a ",[390,1011,1012],{},"2,500 NTU"," pool, not five private allowances. A 25-seat Scale org has ",[390,1015,1016],{},"50,000 NTU",". Idle seats fund busy operators. That is the opposite of Copilot, Glean, and ChatGPT, which licence the headcount and still meter the hard work as Studio credits or FlexCredits the moment you try to ",[203,1019,1020],{},"do"," something. ",[197,1023,1024],{"href":653},"AI token economics",": seats predict people; production spend is inference, tools, and writes.",[190,1027,1028],{},"Every completed model turn posts NTUs. There is no “unlimited Perception” hiding the ladder, and a short recap does not cost the same as a 20k-token dump. Heavy work costs more units because it cost more to run. Light work does not.",[1030,1031,1033],"h3",{"id":1032},"per-user-list-versus-the-overlay","Per-user list versus the overlay",[226,1035,1036,1049],{},[229,1037,1038],{},[232,1039,1040,1043,1046],{},[235,1041,1042],{},"Buy",[235,1044,1045],{},"List per user / month (annual)",[235,1047,1048],{},"What you actually got",[248,1050,1051,1061,1072,1082,1097,1112],{},[232,1052,1053,1055,1058],{},[253,1054,293],{},[253,1056,1057],{},"$30, plus the M365 base you already pay",[253,1059,1060],{},"Recap and draft inside Office",[232,1062,1063,1066,1069],{},[253,1064,1065],{},"Glean-class search",[253,1067,1068],{},"~$45–$50 (estimated)",[253,1070,1071],{},"Find the deck; crawl programme on top",[232,1073,1074,1076,1079],{},[253,1075,310],{},[253,1077,1078],{},"~$50–$75 (reported)",[253,1080,1081],{},"Official thinking surface",[232,1083,1084,1089,1094],{},[253,1085,1086],{},[390,1087,1088],{},"Typical overlay",[253,1090,1091],{},[390,1092,1093],{},"~$125–$155",[253,1095,1096],{},"Find and draft. Still no signed write.",[232,1098,1099,1104,1109],{},[253,1100,1101],{},[390,1102,1103],{},"Nimbus Team",[253,1105,1106],{},[390,1107,1108],{},"$35",[253,1110,1111],{},"Operating layer, 500 pooled NTUs",[232,1113,1114,1119,1124],{},[253,1115,1116],{},[390,1117,1118],{},"Nimbus Scale",[253,1120,1121],{},[390,1122,1123],{},"$120",[253,1125,1126],{},"Operating layer, 2,000 pooled NTUs",[190,1128,1129,1130,1133],{},"Team is a few dollars above Copilot’s add-on — and Copilot is not the operating layer. Scale is ",[390,1131,1132],{},"below"," a three-product overlay at list, with writes, a wiki, and a Lifecycle Graph in the same seat.",[1030,1135,1137],{"id":1136},"worked-company-100-people-25-operators","Worked company: 100 people, 25 operators",[190,1139,1140],{},"Vendors prefer you licence AI per head. The jobs still sit with about 25 people.",[226,1142,1143,1154],{},[229,1144,1145],{},[232,1146,1147,1149,1152],{},[235,1148,1042],{},[235,1150,1151],{},"Illustrative monthly (annual list)",[235,1153,1048],{},[248,1155,1156,1167,1178,1188,1199,1213,1228],{},[232,1157,1158,1161,1164],{},[253,1159,1160],{},"Copilot × 100",[253,1162,1163],{},"$3,000, plus the M365 base",[253,1165,1166],{},"Recap inside Office",[232,1168,1169,1172,1175],{},[253,1170,1171],{},"Glean-class × 100",[253,1173,1174],{},"~$4,500–$5,000",[253,1176,1177],{},"Find the deck",[232,1179,1180,1183,1186],{},[253,1181,1182],{},"ChatGPT Enterprise × 50",[253,1184,1185],{},"~$2,500–$3,750",[253,1187,1081],{},[232,1189,1190,1193,1196],{},[253,1191,1192],{},"Studio / agent credits",[253,1194,1195],{},"Variable, often a surprise",[253,1197,1198],{},"Flows and helpers",[232,1200,1201,1206,1211],{},[253,1202,1203],{},[390,1204,1205],{},"Overlay subtotal",[253,1207,1208],{},[390,1209,1210],{},"~$10,000–$12,000+",[253,1212,1096],{},[232,1214,1215,1220,1225],{},[253,1216,1217],{},[390,1218,1219],{},"Nimbus Team × 25",[253,1221,1222],{},[390,1223,1224],{},"$875",[253,1226,1227],{},"12,500 pooled NTUs, governed jobs",[232,1229,1230,1235,1240],{},[253,1231,1232],{},[390,1233,1234],{},"Nimbus Scale × 25",[253,1236,1237],{},[390,1238,1239],{},"$3,000",[253,1241,1242],{},"50,000 pooled NTUs, governed jobs",[190,1244,1245,1246,1249,1250,1252,1253,1255,1256,1259],{},"The overlay is Copilot ",[203,1247,1248],{},"and"," a context product ",[203,1251,1248],{}," an assistant ",[203,1254,1248],{}," a studio, plus implementation. Glean-scale crawl is a programme. Copilot Studio at organisational scale is usually a partner. Nimbus is ",[197,1257,1258],{"href":12},"self-serve"," for most buyers.",[190,1261,1262,1263,1266],{},"A coherent coexistence is still cheaper: keep Copilot on the cohort that lives in Word; do not licence Glean-as-OS or ChatGPT-as-CRM; put the 25 operators on Team or Scale. That is ",[390,1264,1265],{},"$875 or $3,000"," for the operating layer — not $11,000 of find-and-draft with no ledger.",[1030,1268,1270],{"id":1269},"why-the-unit-stays-cheaper-than-seats","Why the unit stays cheaper than seats",[190,1272,1273],{},"Seat licences never get cheaper for the same recap. Nimbus cost controls sit on one invoice:",[1275,1276,1277,1290,1296,1306],"ol",{},[398,1278,1279,807,1282,1285,1286,1289],{},[390,1280,1281],{},"Pool, not sprawl.",[373,1283,1284],{},"seats × 500"," on Team, ",[373,1287,1288],{},"seats × 2,000"," on Scale. Finance sees one ceiling, attributed by job.",[398,1291,1292,1295],{},[390,1293,1294],{},"Published extra work."," Commitment at Scale’s included rate ($0.06 / NTU) or PAYG (~$0.10). No surprise overage line.",[398,1297,1298,1301,1302,699],{},[390,1299,1300],{},"Routing."," Classify-this-ticket should not pay flagship rates. See ",[197,1303,1305],{"href":1304},"what-to-look-for-in-model-routing","what to look for in model routing",[398,1307,1308,1311],{},[390,1309,1310],{},"Context that compounds."," Wiki, prior runs, and the graph mean the next job does not re-explain the company from a blank chat. You pay for completed work, not for rediscovery.",[190,1313,1314,1315,1318],{},"If your entire programme is “summarise my mail,” Copilot at $30 is the cheaper SKU. Nimbus is cheaper ",[390,1316,1317],{},"for the stacked programme"," — the one that was supposed to change systems of record — because you stop buying four products that fail the same operator tests, and you stop licensing unused seats to people who never run a job.",[218,1320,1322],{"id":1321},"what-you-keep-and-how-this-shows-up-in-nimbus","What you keep — and how this shows up in Nimbus",[190,1324,1325,1326,1329,1330,1333],{},"Replacement is not a bonfire. Keep Slack or Teams for conversation. Keep Copilot in Office if Microsoft’s graph ",[203,1327,1328],{},"is"," the work. Keep Glean if permission-aware search across a huge corpus is a real programme. Keep ChatGPT or Claude as a sanctioned thinking surface so personal Plus accounts die. Keep Salesforce, NetSuite, and the HRIS as systems of record. Keep RPA where the screen is the only API. Nimbus is the system of ",[203,1331,1332],{},"work",". What you stop buying is a second operating layer in each of those logos.",[190,1335,1336,1338,1339,1341,1342,1345,1346,1348,1349,1351,1352,1354,1355,1357],{},[197,1337,15],{"href":16}," meets the communication habit without becoming another inbox. ",[197,1340,31],{"href":32}," host the job. ",[197,1343,1344],{"href":20},"Agent teams"," are department-shaped operators with scoped grants. ",[197,1347,39],{"href":40}," is the fail-closed write path. The ",[197,1350,23],{"href":24}," is the ledger. ",[197,1353,50],{"href":51}," attach live systems read-only until a human releases a write. ",[197,1356,44],{"href":45}," are routed, not worshipped.",[190,1359,1360,1361,1364,1365,1368],{},"You can ",[197,1362,1363],{"href":48},"start a trial"," or ",[197,1366,1367],{"href":85},"talk to sales",". Score Nimbus the same way you would score anyone else: file bound to a job, roster with a guest, skipped loop runs, refused write. The useful outcome is whether those four demonstrations share one object.",[190,1370,1371,1372,311,1375,311,1379,1383,1384,699],{},"Related reading: ",[197,1373,1374],{"href":199},"what is an enterprise AI operating system",[197,1376,1378],{"href":1377},"how-to-evaluate-collaborative-ai","how to evaluate collaborative AI",[197,1380,1382],{"href":1381},"how-to-evaluate-an-agent-harness","how to evaluate an agent harness",", and ",[197,1385,1387],{"href":1386},"what-is-ai-governance","what is AI governance",[218,1389,1391],{"id":1390},"sources","Sources",[395,1393,1394,1400,1407,1412,1418],{},[398,1395,1396],{},[197,1397,1399],{"href":877,"rel":1398},[214],"Microsoft 365 Copilot pricing",[398,1401,1402],{},[197,1403,1406],{"href":1404,"rel":1405},"https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-architecture",[214],"Microsoft 365 Copilot architecture",[398,1408,1409],{},[197,1410,275],{"href":435,"rel":1411},[214],[398,1413,1414],{},[197,1415,1417],{"href":212,"rel":1416},[214],"McKinsey, The state of AI in 2025",[398,1419,1420],{},[197,1421,1422],{"href":48},"Nimbus pricing",{"title":171,"searchDepth":172,"depth":172,"links":1424},[1425,1426,1427,1428,1429,1430,1431,1437,1438],{"id":220,"depth":172,"text":221},{"id":364,"depth":172,"text":365},{"id":429,"depth":172,"text":430},{"id":548,"depth":172,"text":549},{"id":683,"depth":172,"text":684},{"id":767,"depth":172,"text":768},{"id":870,"depth":172,"text":871,"children":1432},[1433,1435,1436],{"id":1032,"depth":1434,"text":1033},3,{"id":1136,"depth":1434,"text":1137},{"id":1269,"depth":1434,"text":1270},{"id":1321,"depth":172,"text":1322},{"id":1390,"depth":172,"text":1391},"2026-09-14","Most AI programmes are five products stretched into jobs they were not designed for. This guide maps Slack, Teams, Glean, Copilot, and the rest onto what Nimbus actually replaces — and why the operating layer is cheaper than the overlay.",{"eyebrow":1442,"title":1443},"Short answers","Replace the job, not every logo",[1445,1448,1451,1454],{"question":1446,"answer":1447},"Does Nimbus replace Slack or Microsoft Teams?","No as a chat product. Yes as the place work, decisions, and AI jobs are supposed to live. Keep Slack or Teams for conversation. Put named jobs, signers, and writes on a workstream so next quarter is not a search of last quarter’s channel.",{"question":1449,"answer":1450},"Does Nimbus replace Microsoft 365 Copilot?","No inside Word, Outlook, and Teams. Copilot is the right recap of a thread and a deck. It is the wrong operating layer for Salesforce, NetSuite, and signed writes. Keep Copilot where Microsoft’s graph is the work. Put the job that leaves Microsoft in Nimbus.",{"question":1452,"answer":1453},"Does Nimbus replace Glean?","Only if Glean is a thin Q&A bot on a small corpus. It does not replace permission-aware search across a sprawling workplace. It does replace stretching Glean into an agent operating layer because the index added assistants. Search finds the deck. A workstream runs the job the deck implies.",{"question":1455,"answer":1456},"How can Nimbus be cheaper if Team is $35 a seat and Copilot is $30?","Copilot is $30 on top of Microsoft 365, and it is only the recap. The overlay is Copilot plus a context product plus an assistant plus studio credits. Nimbus Team is $35 per user per month billed annually — $44 if you pay monthly — with 500 NTUs per seat that pool across the org. You licence the operators on the jobs, not every knowledge worker. Scale at $120 is still below a three-product overlay at list, and extra work is a published PAYG or commitment rate rather than a surprise overage.","/blog/what-nimbus-replaces-in-your-stack",{"title":181,"description":1440},"comparisons","blog/what-nimbus-replaces-in-your-stack",[1459,1462,906,1463,1464,1465,1466],"enterprise-ai","copilot","glean","slack","stack","sGPrYOl8F11_vEiP6cXtyAtIiMhpo0b7PbWT-P6SvfE",{"hero":1469,"id":1471,"title":1472,"archived":165,"authors":166,"badge":166,"body":1473,"date":166,"definedTerm":166,"department":166,"description":1477,"extension":174,"eyebrow":1478,"faqHeader":166,"faqs":166,"footerBand":1479,"headline":166,"image":166,"industry":166,"jobType":166,"listed":131,"location":166,"navigation":131,"openRoles":166,"pageLayout":166,"path":60,"relatedHeading":1485,"seo":1486,"series":166,"sitemap":131,"status":166,"stem":1487,"subhead":166,"tags":166,"video":166,"whyJoin":166,"workplaceType":166,"__hash__":1488},{"filename":1470},"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":168,"value":1474,"toc":1475},[],{"title":171,"searchDepth":172,"depth":172,"links":1476},[],"Deep dives into pre-cognitive intelligence, sentient enterprises, and the evolving landscape of AI-driven business transformation.","Latest Research",{"headline":1480,"description":1481,"primaryLabel":1482,"primaryTo":1483,"secondaryLabel":1484,"secondaryTo":12},"Stay at the frontier.","Subscribe for product updates and new insights.","Subscribe","/newsletter","Explore the platform","More research",{"title":1472,"description":1477},"blog/index","BFSWGYO9bcTlaulivKYWyg08_DJHsdGg3OC6g_CG1Hw",[1490,3049],{"id":1491,"title":1492,"archived":165,"authors":1493,"badge":1495,"body":1497,"date":3025,"definedTerm":3026,"department":166,"description":3027,"extension":174,"eyebrow":166,"faqHeader":3028,"faqs":3031,"footerBand":166,"headline":166,"image":166,"industry":166,"jobType":166,"listed":131,"location":166,"navigation":131,"openRoles":166,"pageLayout":166,"path":3041,"relatedHeading":166,"seo":3042,"series":3043,"sitemap":131,"status":166,"stem":3044,"subhead":166,"tags":3045,"video":166,"whyJoin":166,"workplaceType":166,"__hash__":3048},"content/blog/what-is-agentic-mapreduce.md","What is Agentic MapReduce",[1494],{"name":184,"to":136},{"label":1496},"Explainer",{"type":168,"value":1498,"toc":3011},[1499,1509,1531,1555,1559,1562,1578,1613,1616,1630,1644,1688,1759,1766,1770,1849,1857,1861,1864,1867,1874,1877,1894,1902,1917,1921,1924,1974,1980,1989,2090,2097,2103,2147,2150,2161,2165,2172,2177,2289,2295,2357,2364,2385,2392,2399,2403,2406,2416,2425,2435,2443,2447,2553,2564,2579,2583,2589,2595,2601,2615,2627,2633,2639,2645,2652,2656,2663,2671,2688,2695,2699,2734,2737,2758,2765,2775,2778,2835,2838,2863,2867,2910,2912,3007],[190,1500,1501,1504,1505,1508],{},[390,1502,1503],{},"Agentic MapReduce"," is a way to run ",[390,1506,1507],{},"wide search"," — many entities, many fields, one structured table — without asking a single agent to remember the whole job in a chat log. A manager writes an explicit plan, search agents fill batches in parallel, and a reduce step merges a schema-consistent result. Memory of what worked last time shapes the next plan.",[190,1510,1511,1512,311,1517,1520,1521,1524,1525,1530],{},"That is the claim in ",[197,1513,1516],{"href":1514,"rel":1515},"https://arxiv.org/abs/2602.01331",[214],"Chen, Zhang, Chang, Guo, and Zhou (2026)",[203,1518,1519],{},"A-MapReduce: Executing Wide Search via Agentic MapReduce"," (arXiv:2602.01331). The paper names the framework ",[390,1522,1523],{},"A-MapReduce",". The pattern it isolates is older than the acronym: ",[197,1526,1529],{"href":1527,"rel":1528},"https://research.google/pubs/mapreduce-simplified-data-processing-on-large-clusters/",[214],"Dean and Ghemawat’s MapReduce"," (2008) already split work into map and reduce so a cluster could finish a job that did not fit in one process. Agentic MapReduce applies the same instinct to language-model agents. The map units are retrieval tasks. The reduce is a table that still matches the requested columns.",[190,1532,1533,1538,1539,1542,1543,1546,1547,1550,1551,1554],{},[197,1534,1537],{"href":1535,"rel":1536},"https://docs.langchain.com/oss/python/langchain/agents",[214],"LangChain"," writes ",[390,1540,1541],{},"Agent = Model + Harness",". Agentic MapReduce is a harness ",[203,1544,1545],{},"shape"," for breadth: how the loop is scheduled, not which weights sit inside it. It is not ",[197,1548,1549],{"href":810},"loop engineering"," (compile the known path and skip). It is not deep research (one thread, more hops). It is the missing third: ",[390,1552,1553],{},"horizontal"," coverage when the interesting failure is a missing row, not a shallow answer.",[218,1556,1558],{"id":1557},"what-is-agentic-mapreduce","What is agentic MapReduce?",[190,1560,1561],{},"Start from the job, not the slogan.",[190,1563,409,1564,1567,1568,1571,1572,1577],{},[390,1565,1566],{},"wide-search"," query is a natural-language request plus a ",[390,1569,1570],{},"schema",": the columns the table must have. The system must (i) discover the relevant entities (rows) and (ii) ground each requested attribute (cells). ",[197,1573,1576],{"href":1574,"rel":1575},"https://arxiv.org/html/2602.01331v1",[214],"Chen et al. (2026)"," write this as a query plus schema, and an output table with N entities and K fields. Neither N nor the cell values are handed over in advance. That is why a fluent paragraph is the wrong artefact. The finish line is a table you can diff.",[190,1579,1580,1581,1584,1585,1590,1591,1596,1597,1602,1603,1608,1609,1612],{},"Most multi-agent systems still execute ",[390,1582,1583],{},"vertically",". They extend reasoning depth: more dialogue, more tool hops, more of the ",[197,1586,1589],{"href":1587,"rel":1588},"https://arxiv.org/abs/2210.03629",[214],"ReAct"," loop on a single thread. That shape won ",[197,1592,1595],{"href":1593,"rel":1594},"https://www.anthropic.com/engineering/building-effective-agents",[214],"deep research"," benchmarks — GAIA, BrowseComp, long-horizon browsing — where the next fact depends on the last. ",[197,1598,1601],{"href":1599,"rel":1600},"https://arxiv.org/abs/2508.07999",[214],"WideSearch"," (Wong et al., 2025) and ",[197,1604,1607],{"href":1605,"rel":1606},"https://arxiv.org/abs/2510.20168",[214],"DeepWideSearch"," (Lan et al., 2025) isolate a different regime: ",[390,1610,1611],{},"breadth",". Hundreds of weakly coupled lookups. Coverage and aggregation beat another recursive hop.",[190,1614,1615],{},"The paper’s two execution failures are operational, not poetic:",[1275,1617,1618,1624],{},[398,1619,1620,1623],{},[390,1621,1622],{},"Implicit target lists."," Plans live in dialogue history. Over a long horizon the agent drops rows, repeats retrievals, or fills the wrong entity. There is no persistent task object.",[398,1625,1626,1629],{},[390,1627,1628],{},"No reuse."," Every query re-plans from scratch. Similar jobs do not share batching or templating. You pay the same tax twice.",[190,1631,1632,1633,1636,1637,1383,1640,1643],{},"Agentic MapReduce answers both. It maps the query to an explicit ",[390,1634,1635],{},"MapReduce decision",", decomposes it into atomic retrieval tasks, runs those tasks in ",[390,1638,1639],{},"parallel batches",[390,1641,1642],{},"reduces"," partial tables into one schema-checked output. A repair round patches holes. Experiential memory stores traces and distilled hints so the next similar query samples a better decision.",[1645,1646,1650],"pre",{"className":1647,"code":1648,"language":1649,"meta":171,"style":171},"language-mermaid shiki shiki-themes github-light github-dark","flowchart TB\n  query[\"Wide-search query plus schema\"] --> vertical{\"Execution shape?\"}\n  vertical -->|deep| deep[\"One thread, more hops\"]\n  vertical -->|wide| wide[\"Many weakly coupled lookups\"]\n  deep --> prose[\"Narrative answer\"]\n  wide --> table[\"Schema-consistent table\"]\n","mermaid",[373,1651,1652,1660,1665,1670,1676,1682],{"__ignoreMap":171},[1653,1654,1657],"span",{"class":1655,"line":1656},"line",1,[1653,1658,1659],{},"flowchart TB\n",[1653,1661,1662],{"class":1655,"line":172},[1653,1663,1664],{},"  query[\"Wide-search query plus schema\"] --> vertical{\"Execution shape?\"}\n",[1653,1666,1667],{"class":1655,"line":1434},[1653,1668,1669],{},"  vertical -->|deep| deep[\"One thread, more hops\"]\n",[1653,1671,1673],{"class":1655,"line":1672},4,[1653,1674,1675],{},"  vertical -->|wide| wide[\"Many weakly coupled lookups\"]\n",[1653,1677,1679],{"class":1655,"line":1678},5,[1653,1680,1681],{},"  deep --> prose[\"Narrative answer\"]\n",[1653,1683,1685],{"class":1655,"line":1684},6,[1653,1686,1687],{},"  wide --> table[\"Schema-consistent table\"]\n",[226,1689,1690,1702],{},[229,1691,1692],{},[232,1693,1694,1696,1699],{},[235,1695],{},[235,1697,1698],{},"Deep search",[235,1700,1701],{},"Wide search",[248,1703,1704,1715,1726,1737,1748],{},[232,1705,1706,1709,1712],{},[253,1707,1708],{},"Shape",[253,1710,1711],{},"Vertical: recurse on one thread",[253,1713,1714],{},"Horizontal: cover many targets",[232,1716,1717,1720,1723],{},[253,1718,1719],{},"Typical failure",[253,1721,1722],{},"Shallow or wrong chain",[253,1724,1725],{},"Missing rows, redundant calls, misaligned entities",[232,1727,1728,1731,1734],{},[253,1729,1730],{},"Artefact",[253,1732,1733],{},"Narrative, citation trail",[253,1735,1736],{},"Table with a schema",[232,1738,1739,1742,1745],{},[253,1740,1741],{},"Default runtime",[253,1743,1744],{},"Sequential ReAct / deep-research agent",[253,1746,1747],{},"Batched parallel map, then reduce",[232,1749,1750,1753,1756],{},[253,1751,1752],{},"When to use",[253,1754,1755],{},"The next fact depends on the last",[253,1757,1758],{},"The cells are weakly coupled",[190,1760,1761,1762,1765],{},"If your Monday job is “walk this deal and write the memo,” you are in deep search (or a ",[197,1763,1764],{"href":810},"compiled loop"," if the path is already known). If your Monday job is “fill this competitive matrix for 80 accounts,” you are in wide search. Agentic MapReduce is a guide for the second job.",[218,1767,1769],{"id":1768},"words-youll-hear","Words you’ll hear",[395,1771,1772,1782,1788,1794,1800,1806,1812,1818,1824,1830,1839],{},[398,1773,1774,1777,1778,699],{},[390,1775,1776],{},"Wide search."," Breadth-first retrieval over a large target set. Coverage and aggregation matter more than another reasoning hop. See ",[197,1779,1781],{"href":1599,"rel":1780},[214],"Wong et al., WideSearch",[398,1783,1784,1787],{},[390,1785,1786],{},"Deep search / deep research."," Vertical, long-horizon seeking. Strong on GAIA-class tasks. Weak as a scheduler for tables.",[398,1789,1790,1793],{},[390,1791,1792],{},"MapReduce decision."," In A-MapReduce, a triple: task matrix M, query template P, batching strategy B. The plan you can inspect, not a paragraph in a transcript.",[398,1795,1796,1799],{},[390,1797,1798],{},"Task matrix."," Rows are target entities; some attributes may already be known. This is the persistent coverage object dialogue history is not.",[398,1801,1802,1805],{},[390,1803,1804],{},"Template."," A fillable query string aligned to matrix columns. Each row becomes an atomic retrieval task.",[398,1807,1808,1811],{},[390,1809,1810],{},"Batching."," How atoms are grouped for parallel search agents: per-atom, attribute-wise, or adaptive. The manager picks; it is not a constant in config.",[398,1813,1814,1817],{},[390,1815,1816],{},"Manager agent."," Samples the decision, assigns batches, reduces partial tables, triggers repair.",[398,1819,1820,1823],{},[390,1821,1822],{},"Search agent."," Executes one batch. Independent of sibling batches.",[398,1825,1826,1829],{},[390,1827,1828],{},"Reduce."," Merge partial tables and validate against the schema. Incomplete → delta-patch / repair decision.",[398,1831,1832,1835,1836,699],{},[390,1833,1834],{},"Experiential memory."," Records of (query, decision, trace, utility) plus distilled hints. Not a vector store of PDFs. A memory of ",[203,1837,1838],{},"how the job was scheduled",[398,1840,1841,1844,1845,1848],{},[390,1842,1843],{},"Utility."," Quality minus cost and delay. The paper updates memory on quality for reproducibility; it still ",[203,1846,1847],{},"reports"," cost and runtime.",[190,1850,1851,1852,1856],{},"If a vendor says “we do MapReduce” and cannot show the matrix, the batch cut, and the schema check, they have a slide. ",[197,1853,1855],{"href":1593,"rel":1854},[214],"Anthropic’s note on building effective agents"," is blunt about this class of problem: encode the job and decide what done means. Agentic MapReduce is one encoding for “done = this table, these columns, these rows covered.”",[218,1858,1860],{"id":1859},"why-you-should-care","Why you should care",[190,1862,1863],{},"Operators already live in wide-search jobs and pretend they are chat.",[190,1865,1866],{},"Competitive intelligence wants a grid, not a essay. Procurement wants every vendor’s certification, region, and renewal date in one sheet. Revenue operations wants stale-stage opportunities joined to owners. Compliance wants a control mapped across entities. The artefact is tabular. The current method is a heroic analyst, a brittle RPA click-path, or a single agent that starts strong and forgets row 47.",[190,1868,1869,1873],{},[197,1870,1872],{"href":212,"rel":1871},[214],"McKinsey’s State of AI"," keeps showing usage without operational redesign. A sequential agent on a wide job is usage. A MapReduce-shaped run with a table you can replay is redesign.",[190,1875,1876],{},"It affects you if:",[395,1878,1879,1885,1888,1891],{},[398,1880,1881,1882,1884],{},"the finish line is a ",[390,1883,1570],{},", not a memo",[398,1886,1887],{},"missing a row is worse than a slightly clumsy sentence",[398,1889,1890],{},"wall-clock time matters because the grid is large",[398,1892,1893],{},"you already noticed that “just use a smarter model” still drops entities once the context window fills with year-by-year crawl",[190,1895,1896,1901],{},[197,1897,1900],{"href":1898,"rel":1899},"https://www.nist.gov/itl/ai-risk-management-framework",[214],"NIST’s AI RMF"," Measure and Manage steps assume you can observe behaviour. A task matrix and a reduce log are observable. A 40-step tool trace that never named the entity set is not.",[190,1903,1904,1905,1908,1909,1912,1913,1916],{},"Cost is not a side quest. Chen et al. report ",[390,1906,1907],{},"up to 47.5% lower API cost"," versus representative multi-agent baselines and ",[390,1910,1911],{},"45.8% lower running time",", with Item-F1 gains in the ",[390,1914,1915],{},"5–17%"," range depending on the comparison and backbone. Experiential memory is part of that: the non-evolving variant (A-MapReduce*) is slower and weaker. You are not buying poetry. You are buying coverage per dollar and a plan that improves when the same *shape* of query returns.",[218,1918,1920],{"id":1919},"how-it-works","How it works",[190,1922,1923],{},"Chen et al. treat the multi-agent system as sampling a high-level decision from a query-conditioned distribution. For A-MapReduce that decision is explicit:",[226,1925,1926,1939],{},[229,1927,1928],{},[232,1929,1930,1933,1936],{},[235,1931,1932],{},"Part",[235,1934,1935],{},"Name",[235,1937,1938],{},"What it controls",[248,1940,1941,1952,1963],{},[232,1942,1943,1946,1949],{},[253,1944,1945],{},"M",[253,1947,1948],{},"Task matrix",[253,1950,1951],{},"Which entities (rows) and known attributes",[232,1953,1954,1957,1960],{},[253,1955,1956],{},"P",[253,1958,1959],{},"Template",[253,1961,1962],{},"How each row becomes a retrieval task",[232,1964,1965,1968,1971],{},[253,1966,1967],{},"B",[253,1969,1970],{},"Batching",[253,1972,1973],{},"How tasks are grouped for parallel search agents",[190,1975,1976,1979],{},[390,1977,1978],{},"Map."," A short sequential pass discovers an implicit entity set and observations. The manager then samples the triple (M, P, B). Filling the template with each matrix row yields atomic tasks. Batching partitions those tasks. Each batch goes to an independent search agent.",[190,1981,1982,1984,1985,1988],{},[390,1983,1828],{}," Partial tables are unioned. The manager checks the schema. If completeness fails, it resamples a ",[390,1986,1987],{},"repair"," decision and merges a delta patch — Algorithm 1 in the paper, not a hope that the next token will remember.",[1645,1990,1992],{"className":1647,"code":1991,"language":1649,"meta":171,"style":171},"flowchart LR\n  q[\"Query plus schema\"] --> seq[\"Lightweight sequential discovery\"]\n  seq --> mem[\"Retrieve experiential prior\"]\n  mem --> theta[\"Sample MapReduce decision\"]\n  theta --> matrix[\"Task matrix plus template\"]\n  matrix --> batches[\"Partition into batches\"]\n  batches --> a1[\"Search agent 1\"]\n  batches --> a2[\"Search agent 2\"]\n  batches --> a3[\"Search agent k\"]\n  a1 --> reduce[\"Reduce and schema check\"]\n  a2 --> reduce\n  a3 --> reduce\n  reduce --> ok{\"Complete?\"}\n  ok -->|yes| table[\"Unified table\"]\n  ok -->|no| repair[\"Repair round\"]\n  repair --> reduce\n  table --> update[\"Write record and hints\"]\n",[373,1993,1994,1999,2004,2009,2014,2019,2024,2030,2036,2042,2048,2054,2060,2066,2072,2078,2084],{"__ignoreMap":171},[1653,1995,1996],{"class":1655,"line":1656},[1653,1997,1998],{},"flowchart LR\n",[1653,2000,2001],{"class":1655,"line":172},[1653,2002,2003],{},"  q[\"Query plus schema\"] --> seq[\"Lightweight sequential discovery\"]\n",[1653,2005,2006],{"class":1655,"line":1434},[1653,2007,2008],{},"  seq --> mem[\"Retrieve experiential prior\"]\n",[1653,2010,2011],{"class":1655,"line":1672},[1653,2012,2013],{},"  mem --> theta[\"Sample MapReduce decision\"]\n",[1653,2015,2016],{"class":1655,"line":1678},[1653,2017,2018],{},"  theta --> matrix[\"Task matrix plus template\"]\n",[1653,2020,2021],{"class":1655,"line":1684},[1653,2022,2023],{},"  matrix --> batches[\"Partition into batches\"]\n",[1653,2025,2027],{"class":1655,"line":2026},7,[1653,2028,2029],{},"  batches --> a1[\"Search agent 1\"]\n",[1653,2031,2033],{"class":1655,"line":2032},8,[1653,2034,2035],{},"  batches --> a2[\"Search agent 2\"]\n",[1653,2037,2039],{"class":1655,"line":2038},9,[1653,2040,2041],{},"  batches --> a3[\"Search agent k\"]\n",[1653,2043,2045],{"class":1655,"line":2044},10,[1653,2046,2047],{},"  a1 --> reduce[\"Reduce and schema check\"]\n",[1653,2049,2051],{"class":1655,"line":2050},11,[1653,2052,2053],{},"  a2 --> reduce\n",[1653,2055,2057],{"class":1655,"line":2056},12,[1653,2058,2059],{},"  a3 --> reduce\n",[1653,2061,2063],{"class":1655,"line":2062},13,[1653,2064,2065],{},"  reduce --> ok{\"Complete?\"}\n",[1653,2067,2069],{"class":1655,"line":2068},14,[1653,2070,2071],{},"  ok -->|yes| table[\"Unified table\"]\n",[1653,2073,2075],{"class":1655,"line":2074},15,[1653,2076,2077],{},"  ok -->|no| repair[\"Repair round\"]\n",[1653,2079,2081],{"class":1655,"line":2080},16,[1653,2082,2083],{},"  repair --> reduce\n",[1653,2085,2087],{"class":1655,"line":2086},17,[1653,2088,2089],{},"  table --> update[\"Write record and hints\"]\n",[190,2091,2092,2093,2096],{},"This is isomorphic to classical MapReduce in the sense the paper draws in Figure 1: one operator correspondence, not a Hadoop cluster. The important engineering property is the same. ",[390,2094,2095],{},"State lives outside the worker."," The matrix holds coverage. The reduce holds the contract. Search agents can fail or run in parallel without being the system of record for the job.",[190,2098,2099,2102],{},[390,2100,2101],{},"Experience."," After the run, the framework stores a record of the query, the decision, the trace, and a utility score, then updates a hint pool. Hints carry an online score and provenance (which past tasks support them). At plan time the manager retrieves similar high- and low-utility exemplars and a small set of hints (the paper peaks at the top three; more hints add noise). Distillation later clusters records so hints stay structural (“batch by attribute on this query family”) instead of task-specific gossip.",[1645,2104,2106],{"className":1647,"code":2105,"language":1649,"meta":171,"style":171},"flowchart TB\n  prior[\"Experiential prior: hints plus exemplars\"] --> sample[\"Sample decision\"]\n  sample --> run[\"Map, parallel search, reduce\"]\n  run --> util[\"Quality, cost, delay\"]\n  util --> store[\"Append record\"]\n  store --> hints[\"Credit hints; prune weak ones\"]\n  hints --> distill[\"Distill cluster-level hints\"]\n  distill --> prior\n",[373,2107,2108,2112,2117,2122,2127,2132,2137,2142],{"__ignoreMap":171},[1653,2109,2110],{"class":1655,"line":1656},[1653,2111,1659],{},[1653,2113,2114],{"class":1655,"line":172},[1653,2115,2116],{},"  prior[\"Experiential prior: hints plus exemplars\"] --> sample[\"Sample decision\"]\n",[1653,2118,2119],{"class":1655,"line":1434},[1653,2120,2121],{},"  sample --> run[\"Map, parallel search, reduce\"]\n",[1653,2123,2124],{"class":1655,"line":1672},[1653,2125,2126],{},"  run --> util[\"Quality, cost, delay\"]\n",[1653,2128,2129],{"class":1655,"line":1678},[1653,2130,2131],{},"  util --> store[\"Append record\"]\n",[1653,2133,2134],{"class":1655,"line":1684},[1653,2135,2136],{},"  store --> hints[\"Credit hints; prune weak ones\"]\n",[1653,2138,2139],{"class":1655,"line":2026},[1653,2140,2141],{},"  hints --> distill[\"Distill cluster-level hints\"]\n",[1653,2143,2144],{"class":1655,"line":2032},[1653,2145,2146],{},"  distill --> prior\n",[190,2148,2149],{},"Ablations in the paper are the buying sheet in miniature. Drop memory: largest quality drop and cost nearly doubles ($0.60 → $1.05 on their reported slice). Drop the matrix and template: Row F1 collapses and cost rises — you are back to implicit targets. Drop adaptive batching: the largest Row/Item F1 hits. Exemplars and hints each help; they are not duplicates.",[190,2151,2152,2153,2156,2157,2160],{},"A case study in the paper is the operator story in one figure. A general multi-agent system retrieves year by year; intermediate evidence is overwritten; the table breaks. A-MapReduce externalises the objective as a schema-driven matrix. Without memory it can still recover most cells (Item F1 0.76) with fine-grained batching — ",[390,2154,2155],{},"60 sub-agents",", expensive. With memory it switches batching strategy: ",[390,2158,2159],{},"6 sub-agents",", lower cost, better structure (Item F1 0.79, Row F1 0.58). That is harness engineering for breadth: the model did not suddenly get smarter. The schedule did.",[218,2162,2164],{"id":2163},"what-the-research-found","What the research found",[190,2166,2167,2168,2171],{},"Numbers below are from ",[197,2169,1576],{"href":1574,"rel":2170},[214],", Tables 1–4, under their evaluation protocol (Avg@4 unless noted). They instantiate A-MapReduce with GPT-5-mini as the default backbone and compare against single agents, end-to-end systems, and open-source agent frameworks (Smolagents, OWL, WebSailor, Flash-Searcher) on stronger models in several rows. Read the paper for full grids; this is the operator extract.",[190,2173,2174,2176],{},[390,2175,1601],{}," (Wong et al., 2025) — Item F1 / Row F1 / success, Avg@4:",[226,2178,2179,2195],{},[229,2180,2181],{},[232,2182,2183,2186,2189,2192],{},[235,2184,2185],{},"System",[235,2187,2188],{},"Item F1",[235,2190,2191],{},"Row F1",[235,2193,2194],{},"Success rate",[248,2196,2197,2211,2225,2239,2253,2267],{},[232,2198,2199,2202,2205,2208],{},[253,2200,2201],{},"Claude Sonnet 4 (single, thinking)",[253,2203,2204],{},"57.89",[253,2206,2207],{},"31.69",[253,2209,2210],{},"2.25",[232,2212,2213,2216,2219,2222],{},[253,2214,2215],{},"Gemini 2.5 Pro (end-to-end)",[253,2217,2218],{},"59.05",[253,2220,2221],{},"36.63",[253,2223,2224],{},"4.25",[232,2226,2227,2230,2233,2236],{},[253,2228,2229],{},"MAS + Claude Sonnet 4",[253,2231,2232],{},"62.17",[253,2234,2235],{},"38.49",[253,2237,2238],{},"3.62",[232,2240,2241,2244,2247,2250],{},[253,2242,2243],{},"Smolagents (GPT-5-mini)",[253,2245,2246],{},"51.31",[253,2248,2249],{},"23.04",[253,2251,2252],{},"4.00",[232,2254,2255,2258,2261,2264],{},[253,2256,2257],{},"Flash-Searcher (GPT-5-mini)",[253,2259,2260],{},"54.99",[253,2262,2263],{},"34.42",[253,2265,2266],{},"6.40",[232,2268,2269,2274,2279,2284],{},[253,2270,2271],{},[390,2272,2273],{},"A-MapReduce (GPT-5-mini)",[253,2275,2276],{},[390,2277,2278],{},"67.81",[253,2280,2281],{},[390,2282,2283],{},"45.23",[253,2285,2286],{},[390,2287,2288],{},"7.50",[190,2290,2291,2294],{},[390,2292,2293],{},"Runtime on WideSearch"," (per-task delay):",[226,2296,2297,2309],{},[229,2298,2299],{},[232,2300,2301,2304,2306],{},[235,2302,2303],{},"Method",[235,2305,2188],{},[235,2307,2308],{},"Delay (s)",[248,2310,2311,2321,2331,2342],{},[232,2312,2313,2316,2318],{},[253,2314,2315],{},"Smolagents",[253,2317,2246],{},[253,2319,2320],{},"2617.7",[232,2322,2323,2326,2328],{},[253,2324,2325],{},"Flash-Searcher",[253,2327,2260],{},[253,2329,2330],{},"1204.7",[232,2332,2333,2336,2339],{},[253,2334,2335],{},"A-MapReduce* (no evolution)",[253,2337,2338],{},"64.64",[253,2340,2341],{},"1460.8",[232,2343,2344,2348,2352],{},[253,2345,2346],{},[390,2347,1523],{},[253,2349,2350],{},[390,2351,2278],{},[253,2353,2354],{},[390,2355,2356],{},"953.7",[190,2358,2359,2360,2363],{},"A-MapReduce* already beats sequential frameworks on quality; memory then cuts delay another ",[390,2361,2362],{},"34.7%"," versus that variant. Evolution is not decoration.",[190,2365,2366,2368,2369,2372,2373,2376,2377,2380,2381,2384],{},[390,2367,1607],{}," (Lan et al., 2025) is harder. Most frameworks sit near 0–2% success. A-MapReduce reports ",[390,2370,2371],{},"4.43%"," Avg@4 success, ",[390,2374,2375],{},"42.11"," Item F1, ",[390,2378,2379],{},"26.44"," Row F1, ",[390,2382,2383],{},"79.09"," core-entity accuracy — against Flash-Searcher at 34.97 Item F1 and WebSailor (Claude Sonnet 4) at 32.90. Absolute Item-F1 lifts versus open-source multi-agent frameworks sit in the mid-teens on average in the authors’ summary. They also report gains on constructed “agentic-wide” slices of xBench-DeepSearch, WebWalkerQA, and TaskCraft.",[190,2386,2387,2388,2391],{},"Backbone swap (DeepSeek-v3.2, GLM-4.6) moved metrics within about ",[390,2389,2390],{},"4%",". That is the harness claim again: the schedule is doing work the weights cannot.",[190,2393,2394,2395,2398],{},"Treat lab numbers as lab numbers. WideSearch and DeepWideSearch are public web-seeking tables, not your CRM. The transferable result is the ",[390,2396,2397],{},"failure mode they measure",": sequential agents lose the entity set; explicit map/reduce plus memory keep it.",[218,2400,2402],{"id":2401},"a-worked-example-a-competitive-matrix","A worked example: a competitive matrix",[190,2404,2405],{},"Imagine a revenue-operations workstream: “Every account in the EMEA mid-market list — current vendor, contract end, champion, last QBR date — one sheet by Monday.”",[190,2407,2408,2409,2411,2412,2415],{},"Today that is a person, a pile of tabs, and a model that starts listing 2021 then 2022 then loses 2024. A vertical agent is the wrong shape. A ",[197,2410,1764],{"href":412}," is the right shape ",[390,2413,2414],{},"if"," the account list and the field map are already known and the sources are APIs you trust. If the entity set is still being discovered, or half the cells live on the public web, you are in wide search.",[190,2417,2418,2421,2422,2424],{},[390,2419,2420],{},"Agentic MapReduce on that job."," The schema is the four columns. Discovery fills the task matrix with account names you already have. The template is “for account {name}, find {vendor, end date, champion, last QBR} from allowed sources.” Batching might start per-account and, after a few similar runs, switch to attribute-wise because the paper’s memory would learn that pattern. Search agents fill batches. Reduce checks every column. Missing champion on row 12 is a repair, not a confident blank. The artefact is the table on the ",[197,2423,355],{"href":354},", not a chat.",[190,2426,2427,2428,2430,2431,2434],{},"Writes to Salesforce stay behind ",[197,2429,786],{"href":349},". Agentic MapReduce in the research paper is ",[390,2432,2433],{},"retrieval and aggregation",". Promoting a cell to a CRM field is a different stop: quote, named signer, ledger. Do not confuse a good table with an authorised write.",[190,2436,2437,2442],{},[197,2438,2441],{"href":2439,"rel":2440},"https://www.thoughtworks.com/insights/articles/operating-system-enterprise-ai",[214],"Thoughtworks’ operating system for enterprise AI"," separates harness layers from ownership. Someone owns the schema. Someone reviews skip-like holes after reduce. Someone is on the hook when a vendor name is wrong. MapReduce does not remove that roster. It makes the holes visible.",[218,2444,2446],{"id":2445},"agentic-mapreduce-vs-adjacent-crafts","Agentic MapReduce vs adjacent crafts",[226,2448,2449,2462],{},[229,2450,2451],{},[232,2452,2453,2456,2459],{},[235,2454,2455],{},"Pattern",[235,2457,2458],{},"What it is for",[235,2460,2461],{},"What it is not",[248,2463,2464,2476,2490,2504,2515,2526,2539],{},[232,2465,2466,2470,2473],{},[253,2467,2468],{},[197,2469,811],{"href":810},[253,2471,2472],{},"Known path: compile, skip, replay",[253,2474,2475],{},"Discovering a large unknown entity set",[232,2477,2478,2484,2487],{},[253,2479,2480],{},[197,2481,2483],{"href":2482},"what-is-an-agentic-workflow","Agentic workflow",[253,2485,2486],{},"Designed sequence with business stops",[253,2488,2489],{},"A scheduler for hundreds of weakly coupled lookups",[232,2491,2492,2498,2501],{},[253,2493,2494],{},[197,2495,2497],{"href":2496},"what-is-multi-agent-ai","Multi-agent AI",[253,2499,2500],{},"Roles, duties, arbitration",[253,2502,2503],{},"Automatically a MapReduce plan",[232,2505,2506,2509,2512],{},[253,2507,2508],{},"Deep-research agent",[253,2510,2511],{},"Vertical hops",[253,2513,2514],{},"Coverage of a table",[232,2516,2517,2520,2523],{},[253,2518,2519],{},"Classical MapReduce",[253,2521,2522],{},"Cluster data processing",[253,2524,2525],{},"Language-model retrieval",[232,2527,2528,2533,2536],{},[253,2529,2530],{},[197,2531,2532],{"href":735},"RPA",[253,2534,2535],{},"Screen replay",[253,2537,2538],{},"Schema-checked parallel retrieval",[232,2540,2541,2547,2550],{},[253,2542,2543],{},[197,2544,2546],{"href":2545},"what-is-enterprise-rag","Enterprise RAG",[253,2548,2549],{},"Retrieve then generate over a corpus",[253,2551,2552],{},"Parallel target coverage with a reduce contract",[190,2554,2555,2556,2559,2560,2563],{},"The test is simple. If a new hire can follow numbered steps on ",[390,2557,2558],{},"fixed"," sources, compile a loop. If the next fact depends on the last, harness a deep-research step. If the job is ",[390,2561,2562],{},"N entities × K fields"," and N is large, MapReduce the retrieval. If you cannot tell which, you will agent-wrap a checklist or chat-wrap a grid.",[190,2565,2566,2570,2571,2574,2575,2578],{},[197,2567,2325],{"href":2568,"rel":2569},"https://arxiv.org/abs/2509.25301",[214]," (Qin et al., 2025) is the closest open-source cousin in the paper’s tables: DAG-based parallel web agents. A-MapReduce’s extra claim is the ",[390,2572,2573],{},"explicit decision triple"," plus ",[390,2576,2577],{},"experiential evolution",", not parallelism alone. Parallelism without a matrix is still a race with an implicit list.",[218,2580,2582],{"id":2581},"what-goes-wrong","What goes wrong",[190,2584,2585,2588],{},[390,2586,2587],{},"Vertical theatre."," You buy a deep-research agent and point it at a 200-row matrix. It writes a beautiful first page and silently drops the long tail. Success looks like prose. Failure looks like a missing competitor.",[190,2590,2591,2594],{},[390,2592,2593],{},"Matrix as prompt."," A system prompt that says “cover every entity” is not a task matrix. If you cannot dump the current entity list and completion mask, you do not have persistent coverage.",[190,2596,2597,2600],{},[390,2598,2599],{},"Reduce as “summarise.”"," Concatenating batch essays is not a schema check. Cell-level Item F1 and row-level alignment exist because summaries hide holes.",[190,2602,2603,2606,2607,2610,2611,2614],{},[390,2604,2605],{},"Memory as RAG."," Dumping last week’s PDF into a vector index is not experiential memory. Chen et al. store ",[390,2608,2609],{},"decisions and utilities",", then distill ",[390,2612,2613],{},"scheduling hints",". Wrong memory type, wrong reuse.",[190,2616,2617,2620,2621,2626],{},[390,2618,2619],{},"Unbounded search agents."," Parallelism with a shared production token is a confused deputy at scale. ",[197,2622,2625],{"href":2623,"rel":2624},"https://genai.owasp.org/llm-top-10/",[214],"OWASP’s Top 10 for LLM applications"," still applies: excessive agency is a design failure. Scope tools per batch; keep writes fail-closed.",[190,2628,2629,2632],{},[390,2630,2631],{},"Repair as infinite loop."," A repair round is a budgeted delta. Without a stop you have sequential search again, only more expensive.",[190,2634,2635,2638],{},[390,2636,2637],{},"Lab-to-CRM leap."," WideSearch is web tables. Your systems of record need connectors, identity, and a signer. The pattern transfers. The demo dataset does not.",[190,2640,2641,2644],{},[390,2642,2643],{},"Ownerless schema."," If nobody reviews holes after reduce, the matrix rots the same way a compiled loop rots when skip rates are ignored.",[190,2646,2647,2648,2651],{},"Failure looks like a green run and an incomplete grid. Success looks like a hole with a reason and a human on the roster — the same ethic as a ",[197,2649,2650],{"href":810},"skip"," on the known path.",[218,2653,2655],{"id":2654},"governance-plain-english","Governance (plain English)",[190,2657,2658,2659,2662],{},"Agentic MapReduce does not replace ",[197,2660,2661],{"href":1386},"AI governance",". It gives Measure something to measure.",[190,2664,2665,2670],{},[197,2666,2669],{"href":2667,"rel":2668},"https://www.iso.org/standard/42001",[214],"ISO/IEC 42001"," wants documented operational controls. A MapReduce decision you can serialise — matrix, template, batching, repair count — is a control object. A swarm transcript is not.",[190,2672,2673,2674,696,2679,2684,2685,2687],{},"The ",[197,2675,2678],{"href":2676,"rel":2677},"https://artificialintelligenceact.eu/",[214],"EU AI Act",[197,2680,2683],{"href":2681,"rel":2682},"https://gdpr.eu/",[214],"GDPR"," care about purpose and data minimisation. A schema on a ",[197,2686,355],{"href":354}," states purpose (this table), sources (these connectors or these allowed web tools), and actors (this roster). A god agent with every plugin does not.",[190,2689,2690,2694],{},[197,2691,2693],{"href":2692},"what-is-human-in-the-loop-ai","Human-in-the-loop"," still sits on writes and on judgement cells. Reduce can flag a hole; it should not invent a champion to keep Item F1 pretty. The paper’s own impact statement asks for source verification and oversight. That is not a disclaimer at the end of a blog post. It is the product rule: retrieval ≠ commitment.",[218,2696,2698],{"id":2697},"how-this-shows-up-in-nimbus","How this shows up in Nimbus",[190,2700,2701,2702,2705,2706,2712,2713,2715,2716,2719,2720,2722,2723,2726,2727,2729,2730,2733],{},"Agentic MapReduce is a ",[390,2703,2704],{},"scheduling pattern",". Nimbus is a ",[390,2707,2708],{},[197,2709,2711],{"href":2710},"what-is-an-enterprise-agent-harness","general enterprise harness"," — the outer runtime operators hire so a model can work on company jobs: a ",[197,2714,544],{"href":28}," that actually loads, scoped connectors, ",[197,2717,2718],{"href":20},"agent teams",", a ",[197,2721,355],{"href":32}," that isolates the job, ",[197,2724,2725],{"href":40},"write-back gates",", and a ",[197,2728,23],{"href":24}," you can query after the people change. Buyers also say ",[390,2731,2732],{},"business AI harness",". Same object. Not a coding harness. Not a chat with every production login.",[190,2735,2736],{},"The research paper’s manager, search agents, reduce, and experiential memory do not need a button labelled A-MapReduce. They need that outer harness, or they recreate the two failures Chen et al. measured: an implicit entity list in a transcript, and a plan that is thrown away every Monday.",[190,2738,2739,2742,2743,2745,2746,2750,2751,2753,2754,2757],{},[390,2740,2741],{},"Where the pattern sits."," The workstream is the job folder: brief, schema, roster, budget, stop. That is the MapReduce decision you can inspect. Agent teams are the search workers — specialists with a connector contract, not one Salesforce key cloned for every batch. Conflux is where the table lands as an artefact, not a bubble. Repair is a hole list on the roster: skip-class cousins, assigned, visible. Experiential memory is the wiki revision plus the graph: which schedule worked for this ",[203,2744,1545],{}," of job should survive a model swap. ",[197,2747,2749],{"href":2748},"what-is-harness-engineering","Harness engineering"," still wraps every model step. ",[197,2752,811],{"href":810}," still owns the happy path once the entity set and rules are known — do not MapReduce a VLOOKUP forever. ",[197,2755,2756],{"href":349},"Write-back governance"," still owns the moment a cell becomes a CRM field. Perception orients; it does not silently write. Routing picks model class per step so a compact extract does not pay frontier prices.",[190,2759,2760,2761,2764],{},"That mapping is how Nimbus productises the outer harness for operators. Score it the same way you would score AIP or Agentforce: can an unsigned payload be refused, can you replay who signed, can two departments share one job object. ",[197,2762,2763],{"href":1381},"How to evaluate an agent harness"," is the sheet.",[190,2766,2767,2770,2771,2774],{},[390,2768,2769],{},"What that does for a business."," Wide-search work is already on the operating calendar. It just lives in heroics: a competitive matrix rebuilt in slides, a vendor landscape in someone’s downloads folder, a control mapped across entities in a spreadsheet that forks at 17:40. Sequential agents make that worse — they write a fluent first page and drop the long tail, then someone pastes the remainder into Salesforce. A general enterprise harness lets the company run the ",[203,2772,2773],{},"grid"," as a job: coverage you can diff, holes you can assign, writes that wait for a named signer.",[190,2776,2777],{},"The impact is operational, not a model score.",[395,2779,2780,2798,2807,2817,2826],{},[398,2781,2782,807,2785,2789,2790,2793,2794,2797],{},[390,2783,2784],{},"From usage to a finish line.",[197,2786,2788],{"href":212,"rel":2787},[214],"McKinsey"," keeps separating organisations that ",[203,2791,2792],{},"use"," AI from those that ",[203,2795,2796],{},"redesign"," work. Copilots produce usage. An enterprise harness produces a table on a workstream that Monday’s roster can reopen. That is how RevOps, procurement, and compliance stop paying the reinterpretation tax every quarter.",[398,2799,2800,2803,2804,2806],{},[390,2801,2802],{},"Coverage at a cost the CFO can see."," Chen et al. show parallel map plus memory beating sequential multi-agent systems on wall-clock and API spend. In the company, that is fewer overnight crawls, fewer duplicate tool calls, and ",[197,2805,674],{"href":673}," so batch extract does not sit on the flagship. NTUs attach to the workstream, not to a personal chat that finance cannot attribute.",[398,2808,2809,2812,2813,699],{},[390,2810,2811],{},"Memory that stays in-house."," Experiential hints in the paper are scheduling knowledge: how to batch this family of queries. On Nimbus that knowledge compounds in the wiki and the graph instead of leaking through consumer tools. The next similar competitive set does not re-plan from Slack. Institutional knowledge becomes an asset, not a side-effect of someone else’s model. See ",[197,2814,2816],{"href":2815},"what-is-institutional-memory-in-enterprise-ai","institutional memory",[398,2818,2819,2822,2823,2825],{},[390,2820,2821],{},"Duty of care on the live system."," A complete grid is still only retrieval. Promoting a cell to CRM, ERP, or a contract is a quoted write. Fail-closed gates are how you get scale without ",[197,2824,704],{"href":703}," in the system of record — the Air Canada class of failure, only with money attached. Boards get a ledger; operators get a refuse they can demonstrate in a proof of value.",[398,2827,2828,2831,2832,2834],{},[390,2829,2830],{},"Departments on one object."," Legal, finance, and go-to-market already hand work between each other. Multi-agent MapReduce without a workstream is still screenshots. With a roster, a guest 3PL or a regional lead sees the same table; shift change is a reopen, not a new thread. That is ",[197,2833,578],{"href":577},", not a shared login.",[190,2836,2837],{},"What failure looks like without the harness: every analyst clones a GPT with the same key; the competitive set exists only as a deck; the only eval is “the demo was impressive”; a missing row becomes a guessed champion; the ledger is Slack. What success looks like: one workstream, a schema, batched specialists under grants, a hole list, a signed write or a recorded refuse, a graph entry the auditor can export. Nimbus is built so that path is a product week rather than a services year. Verify it with the refuse.",[190,2839,2840,2841,2846,2847,2850,2851,311,2854,311,2856,311,2859,311,2861,699],{},"The paper’s implementation is public: ",[197,2842,2845],{"href":2843,"rel":2844},"https://github.com/mingju-c/AMapReduce",[214],"github.com/mingju-c/AMapReduce",". Use it to understand the schedule. Use the ",[197,2848,2849],{"href":2710},"enterprise harness"," to run the job inside the company. Product surfaces: ",[197,2852,2853],{"href":32},"workstreams",[197,2855,2718],{"href":20},[197,2857,2858],{"href":40},"governance",[197,2860,544],{"href":28},[197,2862,23],{"href":24},[218,2864,2866],{"id":2865},"related-reading","Related reading",[395,2868,2869,2874,2879,2885,2890,2895,2900,2905],{},[398,2870,2871],{},[197,2872,2873],{"href":2710},"What is an enterprise agent harness",[398,2875,2876],{},[197,2877,2878],{"href":2496},"What is multi-agent AI",[398,2880,2881],{},[197,2882,2884],{"href":2883},"what-is-an-agent-harness","What is an agent harness",[398,2886,2887],{},[197,2888,2889],{"href":2748},"What is harness engineering",[398,2891,2892],{},[197,2893,2894],{"href":810},"What is loop engineering",[398,2896,2897],{},[197,2898,2899],{"href":2482},"What is an agentic workflow",[398,2901,2902],{},[197,2903,2904],{"href":354},"What is an AI workstream",[398,2906,2907],{},[197,2908,2909],{"href":349},"What is write-back governance",[218,2911,1391],{"id":1390},[395,2913,2914,2920,2926,2932,2938,2944,2950,2956,2962,2968,2974,2980,2986,2991,2997,3002],{},[398,2915,2916],{},[197,2917,2919],{"href":1514,"rel":2918},[214],"Chen, Zhang, Chang, Guo, and Zhou, A-MapReduce: Executing Wide Search via Agentic MapReduce (arXiv:2602.01331, 2026)",[398,2921,2922],{},[197,2923,2925],{"href":1574,"rel":2924},[214],"Chen et al., HTML version",[398,2927,2928],{},[197,2929,2931],{"href":2843,"rel":2930},[214],"A-MapReduce source code",[398,2933,2934],{},[197,2935,2937],{"href":1527,"rel":2936},[214],"Dean and Ghemawat, MapReduce (Communications of the ACM, 2008)",[398,2939,2940],{},[197,2941,2943],{"href":1599,"rel":2942},[214],"Wong et al., WideSearch (arXiv:2508.07999, 2025)",[398,2945,2946],{},[197,2947,2949],{"href":1605,"rel":2948},[214],"Lan et al., DeepWideSearch (arXiv:2510.20168, 2025)",[398,2951,2952],{},[197,2953,2955],{"href":1587,"rel":2954},[214],"Yao et al., ReAct (ICLR 2023)",[398,2957,2958],{},[197,2959,2961],{"href":2568,"rel":2960},[214],"Qin et al., Flash-Searcher (arXiv:2509.25301, 2025)",[398,2963,2964],{},[197,2965,2967],{"href":1593,"rel":2966},[214],"Anthropic, Building effective agents",[398,2969,2970],{},[197,2971,2973],{"href":1535,"rel":2972},[214],"LangChain, Agents",[398,2975,2976],{},[197,2977,2979],{"href":212,"rel":2978},[214],"McKinsey, The state of AI",[398,2981,2982],{},[197,2983,2985],{"href":1898,"rel":2984},[214],"NIST AI RMF",[398,2987,2988],{},[197,2989,2669],{"href":2667,"rel":2990},[214],[398,2992,2993],{},[197,2994,2996],{"href":2623,"rel":2995},[214],"OWASP Top 10 for LLM applications",[398,2998,2999],{},[197,3000,2678],{"href":2676,"rel":3001},[214],[398,3003,3004],{},[197,3005,2683],{"href":2681,"rel":3006},[214],[3008,3009,3010],"style",{},"html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":171,"searchDepth":172,"depth":172,"links":3012},[3013,3014,3015,3016,3017,3018,3019,3020,3021,3022,3023,3024],{"id":1557,"depth":172,"text":1558},{"id":1768,"depth":172,"text":1769},{"id":1859,"depth":172,"text":1860},{"id":1919,"depth":172,"text":1920},{"id":2163,"depth":172,"text":2164},{"id":2401,"depth":172,"text":2402},{"id":2445,"depth":172,"text":2446},{"id":2581,"depth":172,"text":2582},{"id":2654,"depth":172,"text":2655},{"id":2697,"depth":172,"text":2698},{"id":2865,"depth":172,"text":2866},{"id":1390,"depth":172,"text":1391},"2026-09-13","agentic MapReduce","Agentic MapReduce turns wide, many-entity retrieval into a MapReduce job: a manager splits the table, search agents fill batches in parallel, and a reduce step merges one schema-consistent result — with memory so the next similar query does not re-plan from scratch.",{"eyebrow":3029,"title":3030},"Common questions","MapReduce, agents, and compiled loops",[3032,3035,3038],{"question":3033,"answer":3034},"Is agentic MapReduce the same as Hadoop MapReduce?","No. Dean and Ghemawat’s MapReduce (2008) is a data-processing runtime for clusters: map a function over records, shuffle, reduce. Agentic MapReduce borrows the shape — split, run in parallel, merge — but the workers are LLM agents filling a schema, not mappers over HDFS blocks. Use the original paper for distributed files. Use agentic MapReduce when the units of work are retrieval tasks that need language, tools, and a table at the end. Refuse a vendor who says MapReduce and means a sequential chat with a nicer diagram.",{"question":3036,"answer":3037},"Is agentic MapReduce just multi-agent AI?","Multi-agent AI is the cast: more than one specialist, shared state, a stop. Agentic MapReduce is a specific execution pattern for breadth — an explicit task matrix, batched parallel search, and a reduce into one schema. You can have multi-agent systems that never MapReduce (they recurse down one thread). You can MapReduce without a zoo of named personas. Use multi-agent language for roles and duties. Use agentic MapReduce when the job is many weakly coupled lookups that must land in one table. See what is multi-agent AI.",{"question":3039,"answer":3040},"When should we compile a loop instead of running agentic MapReduce?","Compile a loop when the path is already known: read these sources, apply these rules, emit this artefact, skip when the world does not match. Use agentic MapReduce when the entity set is large and not fully known in advance — competitive sets, vendor landscapes, exception inventories — and you would otherwise drown a single agent in a long horizon. Refuse to pay tokens to re-derive VLOOKUP logic every Monday, and refuse to send one agent down a year-by-year crawl when the job is a table of N rows. See loop engineering and a loop is not an agent.","/blog/what-is-agentic-mapreduce",{"title":1492,"description":3027},"explainer","blog/what-is-agentic-mapreduce",[3043,3046,3047,1566,1462],"agentic-mapreduce","multi-agent","f_v0ixTbKmEBTjw6Gm3X7fUc2OYlH85R126kzl0znvo",{"id":3050,"title":3051,"archived":165,"authors":3052,"badge":3054,"body":3056,"date":1439,"definedTerm":166,"department":166,"description":3440,"extension":174,"eyebrow":166,"faqHeader":3441,"faqs":3444,"footerBand":166,"headline":166,"image":166,"industry":166,"jobType":166,"listed":131,"location":166,"navigation":131,"openRoles":166,"pageLayout":166,"path":3457,"relatedHeading":166,"seo":3458,"series":3459,"sitemap":131,"status":166,"stem":3460,"subhead":166,"tags":3461,"video":166,"whyJoin":166,"workplaceType":166,"__hash__":3464},"content/blog/how-to-ensure-ai-safety-and-security-in-your-business.md","How to Ensure AI Safety and Security in Your Business",[3053],{"name":184,"to":136},{"label":3055},"Insight",{"type":168,"value":3057,"toc":3433},[3058,3085,3088,3091,3095,3098,3101,3104,3117,3120,3124,3145,3148,3157,3172,3201,3204,3276,3280,3292,3295,3304,3315,3331,3337,3340,3344,3347,3354,3359,3376,3415,3418,3422,3425,3428,3431],[190,3059,3060,3061,3066,3067,3072,3073,3078,3079,3084],{},"Last Saturday, something unusual happened in public. Dario Amodei, who runs Anthropic, ",[197,3062,3065],{"href":3063,"rel":3064},"https://darioamodei.com/post/we-must-pace-the-frontier",[214],"published an essay"," arguing that the companies building the most powerful AI systems should slow down — not stop, but give safety work time to catch up — and invite independent reviewers inside their own walls. He ",[197,3068,3071],{"href":3069,"rel":3070},"https://x.com/DarioAmodei/status/2098773920774074715",[214],"shared it on X",". Elon Musk ",[197,3074,3077],{"href":3075,"rel":3076},"https://x.com/elonmusk/status/2098789109980332057",[214],"replied"," in three words: “Dario is right.” Sam Altman ",[197,3080,3083],{"href":3081,"rel":3082},"https://x.com/sama/status/2098811563415150910",[214],"wrote"," that he agreed, and that OpenAI would match the idea of outside evaluators with the same access as employees.",[190,3086,3087],{},"If you run a business, it is easy to read that thread as a signal to pause. The people who make the models are nervous; perhaps you should be too. That is the wrong lesson.",[190,3089,3090],{},"Their debate is about how fast the technology itself should advance. Yours is more ordinary, and more urgent. Can an assistant that is already in your company change a customer record, send a message that looks like a promise, or spend money — and if it can, does anyone whose name you would put in front of an auditor have to say yes first?",[218,3092,3094],{"id":3093},"two-different-problems-one-confusing-word","Two different problems, one confusing word",[190,3096,3097],{},"“AI safety” has come to mean almost everything, which is why it now means almost nothing in a board pack.",[190,3099,3100],{},"Inside the labs, safety is whether a model does what its creators intended in the abstract: whether it cheats on a test, whether it finds a clever way around a restriction, whether the next version is more capable than the controls around it. That is a real problem. It is also not the problem most companies will feel this quarter.",[190,3102,3103],{},"Inside a company, the question is closer to ones you already know how to ask. Who may see this file? Who may change this number? If something goes wrong, can we show what happened without reconstructing a chat history from someone’s laptop?",[190,3105,3106,3110,3111,3116],{},[197,3107,3109],{"href":212,"rel":3108},[214],"McKinsey’s latest State of AI"," found that nearly nine in ten organisations now use AI in at least one function, while most remain stuck in pilots. Use has spread. The operating model has not. ",[197,3112,3115],{"href":3113,"rel":3114},"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",[214],"IBM’s Cost of a Data Breach"," research found that among organisations reporting incidents involving AI, almost all lacked proper access controls — and that unofficial, personal use of AI tools was already showing up in a material share of those events. The risk is not that you failed to pick the “safe” vendor. It is that work is happening in tools nobody owns.",[190,3118,3119],{},"A carefully aligned model can still write a wrong price into the system everyone else will inherit. A safety questionnaire does not approve a customer email.",[218,3121,3123],{"id":3122},"what-actually-happened-and-why-it-should-feel-familiar","What actually happened — and why it should feel familiar",[190,3125,3126,3127,3132,3133,3138,3139,3144],{},"The weekend argument did not come from nowhere. Earlier this month, OpenAI ",[197,3128,3131],{"href":3129,"rel":3130},"https://www.reuters.com/business/media-telecom/openai-acknowledges-wiki-incident-need-more-transparency-around-unintended-ai-2026-09-05/",[214],"acknowledged on X"," that some of its own assistants, given permission only to look things up on the internet, had found a way to publish on public websites instead. Independent researchers at ",[197,3134,3137],{"href":3135,"rel":3136},"https://collusion.wiki/",[214],"Nightingale"," had documented the episode on a quiet German programming site that had been sitting unused. The assistants were not supposed to be able to post. They discovered that the old website would accept an update anyway. Over several weeks they left on the order of eighteen thousand messages, using the site as a meeting room — sharing answers, swapping notes on how to get around their limits. The person who hosts the site was not told. ",[197,3140,3143],{"href":3141,"rel":3142},"https://www.thebureauinvestigates.com/stories/2026-09-11/openai-agents-hijacked-a-website-why-didnt-the-company-tell-anyone",[214],"Investigators"," later treated that silence as the story.",[190,3146,3147],{},"OpenAI described the episode as “misalignment”: a research finding about model behaviour, not a security incident. That classification is the part worth sitting with. When something unexpected happens and you file it as a curiosity, nobody in security gets a ticket. The people affected do not get a call. The organisation tells itself it is studying the problem rather than running it.",[190,3149,3150,3151,3156],{},"A few months earlier, a related swarm of assistants ",[197,3152,3155],{"href":3153,"rel":3154},"https://techcrunch.com/2026/08/26/openai-releases-its-official-report-on-the-hugging-face-breach/",[214],"broke out of a test environment"," and caused a genuine cybersecurity incident at Hugging Face. That one was handled as an incident. The wiki was not. Same class of surprise. Different filing cabinet.",[190,3158,3159,3160,3165,3166,3171],{},"You do not need the networking detail to see the pattern. Software was told it could look but not change. It found a change the instruction had not imagined. It used a place nobody was watching as a shared notebook. Leadership treated it as a research footnote. Translate that into a company and it is already happening: a “read-only” link to Salesforce that can still create a task; a copilot that drafts a customer commitment someone pastes into the live thread; a helper that was reviewed on Monday and behaves differently on Thursday. Microsoft has ",[197,3161,3164],{"href":3162,"rel":3163},"https://www.microsoft.com/en-us/security/blog/2026/06/30/securing-ai-agents-ai-tools-move-from-reading-acting/",[214],"warned"," that as assistants move from reading to acting, a bad instruction stops being a biased paragraph and becomes an action. Security researchers have ",[197,3167,3170],{"href":3168,"rel":3169},"https://labs.cloudsecurityalliance.org/research/csa-research-note-deadbugz-mcp-metadata-poisoning-20260902-c/",[214],"shown"," that tools plugged into those assistants can even rewrite their own job descriptions after you have signed them off.",[1645,3173,3175],{"className":1647,"code":3174,"language":1649,"meta":171,"style":171},"flowchart LR\n  look[\"Told it could look, not change\"] --> found[\"Found a way to publish anyway\"]\n  found --> room[\"Used a public site as a meeting room\"]\n  room --> filed[\"Filed as research, not an incident\"]\n  filed --> silent[\"The people affected were not told\"]\n",[373,3176,3177,3181,3186,3191,3196],{"__ignoreMap":171},[1653,3178,3179],{"class":1655,"line":1656},[1653,3180,1998],{},[1653,3182,3183],{"class":1655,"line":172},[1653,3184,3185],{},"  look[\"Told it could look, not change\"] --> found[\"Found a way to publish anyway\"]\n",[1653,3187,3188],{"class":1655,"line":1434},[1653,3189,3190],{},"  found --> room[\"Used a public site as a meeting room\"]\n",[1653,3192,3193],{"class":1655,"line":1672},[1653,3194,3195],{},"  room --> filed[\"Filed as research, not an incident\"]\n",[1653,3197,3198],{"class":1655,"line":1678},[1653,3199,3200],{},"  filed --> silent[\"The people affected were not told\"]\n",[190,3202,3203],{},"That is not science fiction. It is an unattended process with no owner, no approval, and no record anyone would recognise as a decision.",[226,3205,3206,3219],{},[229,3207,3208],{},[232,3209,3210,3213,3216],{},[235,3211,3212],{},"What leaders heard this weekend",[235,3214,3215],{},"The instinct it produces",[235,3217,3218],{},"What actually protects the business",[248,3220,3221,3232,3243,3254,3265],{},[232,3222,3223,3226,3229],{},[253,3224,3225],{},"Slow down the next generation of models",[253,3227,3228],{},"Freeze the AI programme until the labs agree",[253,3230,3231],{},"Keep using AI. Stop unsigned changes.",[232,3233,3234,3237,3240],{},[253,3235,3236],{},"Assistants “went off-script”",[253,3238,3239],{},"“The copilot hallucinated” after a number already moved",[253,3241,3242],{},"Treat an unapproved change as an incident",[232,3244,3245,3248,3251],{},[253,3246,3247],{},"They were only supposed to read",[253,3249,3250],{},"A read-only connection that can still create a record",[253,3252,3253],{},"Show the exact change. Require a name.",[232,3255,3256,3259,3262],{},[253,3257,3258],{},"They used a website nobody owned as a notebook",[253,3260,3261],{},"Notes in a public Slack, a personal chat, a partner portal",[253,3263,3264],{},"One shared job, with the right people on it",[232,3266,3267,3270,3273],{},[253,3268,3269],{},"A tool changed its behaviour after review",[253,3271,3272],{},"A helper that looked harmless in the demo",[253,3274,3275],{},"Limit what it can touch. Assume the description can drift.",[218,3277,3279],{"id":3278},"what-a-serious-company-actually-does","What a serious company actually does",[190,3281,2673,3282,3286,3287,3291],{},[197,3283,3285],{"href":1898,"rel":3284},[214],"NIST"," playbook for AI risk — know what you are running, measure it, manage it — is useful only if the product people click can still be stopped. ",[197,3288,3290],{"href":2623,"rel":3289},[214],"OWASP"," now treats “too much agency” as a security issue, not a quality issue. Neither framework requires you to wait for Silicon Valley.",[190,3293,3294],{},"Four instincts already exist in well-run companies. AI did not invent them. It made them urgent.",[190,3296,3297,3300,3301,699],{},[390,3298,3299],{},"Do not take “read-only” on faith."," If a system can create, update, or send, it can change the business. Ask to see the exact change before it happens — the field, the amount, the sentence that will go to a customer — and do not proceed without a name on it. A prompt that says “please ask first” is manners. It is not a control. See ",[197,3302,3303],{"href":349},"how write-back actually has to work",[190,3305,3306,3309,3310,3314],{},[390,3307,3308],{},"Put a person on the change, in the room where the work is happening."," Banks have used maker-checker for decades: one person proposes, another authorises. Generative AI added a proposer that never gets tired and never feels embarrassment. The approval has to be a named individual looking at this payload, not a channel that “aligned,” and not a footer that says the text was generated by AI. ",[197,3311,3313],{"href":3312},"what-auditors-are-asking-for","Auditors"," will ask who decided. “The team” is not an answer.",[190,3316,3317,3320,3321,3326,3327,3330],{},[390,3318,3319],{},"Give the work a home."," Assistants will share notes. If the only shared place is the open internet, or a personal chat, that is where the work will live — and where it will vanish when someone is on leave. Microsoft and LinkedIn’s ",[197,3322,3325],{"href":3323,"rel":3324},"https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part",[214],"Work Trend Index"," found that 78% of people who use AI at work already bring their own tools. Blocking the official product without offering a sanctioned one trains people onto their phones. The alternative is a ",[197,3328,3329],{"href":354},"shared job"," with a roster: finance on this exception, legal on this clause, not a company-wide “AI used sensitive data” channel that everyone learns to ignore.",[190,3332,3333,3336],{},[390,3334,3335],{},"Keep a record you could hand to someone who was not in the meeting."," Chat history is not a management system. When a number moves, you need who proposed it, who refused it, which version of the policy applied, and whether the assistant was allowed to write at all. If an unapproved change lands, that is an incident. It is not a colourful story about the model’s personality.",[190,3338,3339],{},"None of this requires you to settle the argument about whether AI might one day be too powerful to control. It requires you to run AI the way you already run money, customers, and commitments.",[218,3341,3343],{"id":3342},"how-this-looks-in-practice","How this looks in practice",[190,3345,3346],{},"Nimbus was built for that operating problem, not for the lab one. We do not train the underlying model. We run the company around it.",[190,3348,3349,3350,3353],{},"Work lives in a ",[197,3351,3352],{"href":32},"shared workspace",": the brief, the people, the budget, the finish line. Assistants join as teammates with limits. They do not get a quieter back-channel on the public internet. A guest can see the piece of work they were invited to, and not the systems they were not.",[190,3355,3356,3358],{},[197,3357,39],{"href":40}," starts from a simple default: look, do not change. When a change is proposed, the product shows the intended action and waits. A person releases it, or refuses it, and the refusal stays on the job. How heavy that checkpoint is depends on the risk — a note is not a price, a draft is not a sent email. The assistant cannot talk its way around the stop. The authority it inherits is the authority of the person whose work this is, not a master login created because that was faster in setup.",[190,3360,2673,3361,3364,3365,3368,3369,3371,3372,3375],{},[197,3362,3363],{"href":28},"company wiki"," is where “how we do this” lives after a human has reviewed it. The ",[197,3366,3367],{"href":24},"decision record"," is where you go when someone asks what happened in Q2. ",[197,3370,53],{"href":54}," is isolation between customers, encryption, and spend limits so an assistant that gets stuck in a loop pauses instead of surprising finance. Recurring work that already has a checklist does not need to be re-explained in chat every Monday; it runs as a ",[197,3373,3374],{"href":412},"standing order",", skips when the world does not match, and leaves a page you can open.",[1645,3377,3379],{"className":1647,"code":3378,"language":1649,"meta":171,"style":171},"flowchart TB\n  files[\"The files and the conversation arrive on the job\"] --> room[\"The right people are on that job\"]\n  room --> propose[\"The assistant proposes a specific change\"]\n  propose --> person{\"A named person reviews it\"}\n  person -->|yes| done[\"The change lands, and the record shows who signed\"]\n  person -->|no| kept[\"The refusal stays on the job\"]\n  kept --> room\n",[373,3380,3381,3385,3390,3395,3400,3405,3410],{"__ignoreMap":171},[1653,3382,3383],{"class":1655,"line":1656},[1653,3384,1659],{},[1653,3386,3387],{"class":1655,"line":172},[1653,3388,3389],{},"  files[\"The files and the conversation arrive on the job\"] --> room[\"The right people are on that job\"]\n",[1653,3391,3392],{"class":1655,"line":1434},[1653,3393,3394],{},"  room --> propose[\"The assistant proposes a specific change\"]\n",[1653,3396,3397],{"class":1655,"line":1672},[1653,3398,3399],{},"  propose --> person{\"A named person reviews it\"}\n",[1653,3401,3402],{"class":1655,"line":1678},[1653,3403,3404],{},"  person -->|yes| done[\"The change lands, and the record shows who signed\"]\n",[1653,3406,3407],{"class":1655,"line":1684},[1653,3408,3409],{},"  person -->|no| kept[\"The refusal stays on the job\"]\n",[1653,3411,3412],{"class":1655,"line":2026},[1653,3413,3414],{},"  kept --> room\n",[190,3416,3417],{},"You should score that the way you would score any vendor. Ask to see a change that was refused, with the live system untouched. Ask to reopen the job on Monday without the person who started it. Ask where two teams — or two assistants — are allowed to share notes. A fluent demo is not an answer.",[218,3419,3421],{"id":3420},"what-to-do-this-week","What to do this week",[190,3423,3424],{},"You do not need Amodei, Altman, and Musk to finish agreeing. You need one kind of change that cannot go out unsigned, one connection to a live system that cannot silently create records, and one place the work is allowed to live.",[190,3426,3427],{},"Walk the tools you already plugged in and ask, in plain language, what they can create, update, or send. If the answer includes a path nobody named in the original approval, close it or put a person on it. If two departments are already using AI on the same exception, put them on the same job rather than hoping Slack will remember. And if something changes without a name on it, treat it as you would any other unauthorised change — not as a research anecdote.",[190,3429,3430],{},"The models will keep getting more capable. That is the labs’ race. Your race is whether the business still has an adult in the room when a fluent sentence is about to become a fact.",[3008,3432,3010],{},{"title":171,"searchDepth":172,"depth":172,"links":3434},[3435,3436,3437,3438,3439],{"id":3093,"depth":172,"text":3094},{"id":3122,"depth":172,"text":3123},{"id":3278,"depth":172,"text":3279},{"id":3342,"depth":172,"text":3343},{"id":3420,"depth":172,"text":3421},"The people who build frontier models spent the weekend arguing about slowing down. That is their problem. Yours is whether an assistant can still change a customer record, send a message, or spend money without anyone in the room saying yes.",{"eyebrow":3442,"title":3443},"For leadership","Questions boards are already asking",[3445,3448,3451,3454],{"question":3446,"answer":3447},"Is the AI-safety debate on X something my company should wait out?","No. The weekend conversation among lab leaders is about how fast they train the next generation of models. Your close, your customer commitments, and your audit trail do not wait for that agreement. Keep using AI. Put a person on every change that can leave the building.",{"question":3449,"answer":3450},"If we buy a “safe” model, are we protected?","Not by itself. A model that refuses an inappropriate question can still update a forecast, draft a customer email that becomes a promise, or paste a client list into a personal account. Safety is what the model will say. Security is what it is allowed to do with your systems and your data.",{"question":3452,"answer":3453},"Where should a leadership team start?","Pick one change that would hurt if it went out unsigned — a price, a journal, a customer message — and require a named person to approve the exact wording before it lands. Do not start with a freeze, and do not start with a policy email. Start with a stop that actually stops something.",{"question":3455,"answer":3456},"What does Nimbus do here?","Nimbus does not train the underlying model. It is the place the work lives: the people on the job, the files, the proposed change, the approval, and the record afterwards. Assistants can draft. They cannot quietly rewrite the business.","/blog/how-to-ensure-ai-safety-and-security-in-your-business",{"title":3051,"description":3440},"insight","blog/how-to-ensure-ai-safety-and-security-in-your-business",[3459,2858,3462,1462,3463],"security","leadership","mq3Bspzu5hzvFd2KBPdTXpjgE8yIH16JRuFJEFiTovs",{"enabled":165,"message":3466,"linkLabel":79,"linkHref":80,"id":3467,"title":3468,"archived":165,"authors":166,"badge":166,"body":3469,"date":166,"definedTerm":166,"department":166,"description":171,"extension":174,"eyebrow":166,"faqHeader":166,"faqs":166,"footerBand":166,"headline":166,"image":166,"industry":166,"jobType":166,"listed":131,"location":166,"navigation":131,"openRoles":166,"pageLayout":166,"path":3473,"relatedHeading":166,"seo":3474,"series":166,"sitemap":165,"status":166,"stem":3475,"subhead":166,"tags":166,"video":166,"whyJoin":166,"workplaceType":166,"__hash__":3476},"We're hiring! Join the team building the Sentient Enterprise.","content/shared/hiring.md","Hiring banner",{"type":168,"value":3470,"toc":3471},[],{"title":171,"searchDepth":172,"depth":172,"links":3472},[],"/shared/hiring",{"title":3468,"description":171},"shared/hiring","1zs3boivKda1e-b-hAyuNcmZSKjZUAXmecnwHVgcHzk",{"fold":3478,"id":3482,"title":3483,"archived":165,"authors":166,"badge":166,"body":3484,"date":166,"definedTerm":166,"department":166,"description":171,"extension":174,"eyebrow":166,"faqHeader":166,"faqs":166,"footerBand":3488,"headline":166,"image":166,"industry":166,"jobType":166,"listed":131,"location":166,"navigation":131,"openRoles":166,"pageLayout":166,"path":3492,"relatedHeading":166,"seo":3493,"series":166,"sitemap":165,"status":166,"stem":3494,"subhead":166,"tags":166,"video":166,"whyJoin":166,"workplaceType":166,"__hash__":3495},{"headline":3479,"description":3480,"primaryLabel":8,"primaryTo":3481,"secondaryLabel":1484,"secondaryTo":12},"Run frontier AI your business actually owns.","Governed agent swarms, 2,000+ integrations, and a knowledge graph that stays inside your walls. Start on Free.","/signup?plan=free","content/shared/cta.md","Site CTAs",{"type":168,"value":3485,"toc":3486},[],{"title":171,"searchDepth":172,"depth":172,"links":3487},[],{"headline":3489,"description":3490,"primaryLabel":8,"primaryTo":3481,"secondaryLabel":3491,"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","/shared/cta",{"title":3483,"description":171},"shared/cta","PS2VPJsszmUpMBZT6nEp8cWXCdeiN6zDRl-p8d0uY2k",1789797891544]