[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-post-\u002Fnews\u002Fpentagon-100000-ai-agents-vibe-coding-genai-mil-google-gemini-april-2026":3,"news-related-en-\u002Fnews\u002Fpentagon-100000-ai-agents-vibe-coding-genai-mil-google-gemini-april-2026":424,"hreflang-pentagon-100000-ai-agents-vibe-coding-genai-mil-google-gemini-april-2026":444},{"doc":4,"redirect":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":9,"category":401,"date":402,"description":403,"extension":404,"image":405,"meta":406,"navigation":407,"path":408,"seo":409,"seoDescription":410,"seoTitle":411,"stem":412,"tags":413,"translationKey":422,"updatedAt":8,"__hash__":423},"news_en\u002Fnews\u002Fpentagon-100000-ai-agents-vibe-coding-genai-mil-google-gemini-april-2026.md","Pentagon: 100,000 vibe-coded AI agents in five weeks on GenAI.mil with Gemini 3.1 Pro","Paul Lefizelier",null,{"type":10,"value":11,"toc":381},"minimark",[12,45,50,64,75,79,88,100,165,169,192,199,203,206,211,221,225,232,236,243,247,250,254,266,270,281,285,301,305,308,340,344,351,364,367],[13,14,15,16,20,21,24,25,28,29,32,33,36,37,40,41,44],"p",{},"On April 28, 2026, at the ",[17,18,19],"strong",{},"Box Federal Summit",", the ",[17,22,23],{},"Pentagon"," confirmed a number few private companies can match: ",[17,26,27],{},"more than 103,000 AI agents built"," and ",[17,30,31],{},"1.1 million agent sessions logged"," on ",[17,34,35],{},"GenAI.mil",", its internal generative platform — in less than five weeks after ",[17,38,39],{},"Google's Agent Designer"," went live. Officers, analysts and civilians at the ",[17,42,43],{},"Department of Defense"," are vibe coding at scale to automate their workflows, from document triage to operational reporting.",[46,47,49],"h2",{"id":48},"genaimil-the-platform-that-changed-the-game","GenAI.mil, the platform that changed the game",[13,51,52,53,56,57,28,60,63],{},"Launched on ",[17,54,55],{},"December 9, 2025",", GenAI.mil hit ",[17,58,59],{},"500,000 users in a week",[17,61,62],{},"1 million users in a month",", with zero latency, zero downtime. It's one of the fastest enterprise AI adoptions ever observed, public or private.",[13,65,66,67,70,71,74],{},"The architecture is mixed: models from ",[17,68,69],{},"Anthropic, OpenAI, Meta and now Google Gemini 3.1 Pro",", integrated behind an in-house broker. Users access the entire stack through a single interface. And since ",[17,72,73],{},"Agent Designer"," joined the catalog, usage spike has shifted to agent creation.",[46,76,78],{"id":77},"vibe-coding-enters-the-dod","\"Vibe coding\" enters the DoD",[13,80,81,82,87],{},"The mechanism is exactly what consumer developers have known for months through ",[83,84,86],"a",{"href":85},"\u002Fblog\u002Fno-code-vs-vibecoding-vs-classic-dev-startup-guide","Lovable, Bolt and Cursor",". An officer opens Agent Designer, describes in natural language what they want to accomplish (\"build me an agent that compiles every morning the last 24h of patrol reports into a SITREP-formatted brief\"), and the tool generates the agent autonomously.",[13,89,90,91,95,96,99],{},"No syntax, no SDK to learn, no tickets submitted to a central IT team. It's the enterprise embodiment of what ",[83,92,94],{"href":93},"\u002Fnews\u002Fcursor-3-agent-first-interface-parallel-agents-vibe-coding-2026","Cursor 3 and its parallel agent interface"," popularized for civilian developers — applied to a public of ",[17,97,98],{},"non-coders inside government",".",[101,102,103,116],"table",{},[104,105,106],"thead",{},[107,108,109,113],"tr",{},[110,111,112],"th",{},"GenAI.mil metric",[110,114,115],{},"Value",[117,118,119,127,135,145,155],"tbody",{},[107,120,121,125],{},[122,123,124],"td",{},"Launch",[122,126,55],{},[107,128,129,132],{},[122,130,131],{},"Users (D+30)",[122,133,134],{},"1,000,000+",[107,136,137,140],{},[122,138,139],{},"Agents built",[122,141,142],{},[17,143,144],{},"103,000+",[107,146,147,150],{},[122,148,149],{},"Agent sessions",[122,151,152],{},[17,153,154],{},"1.1 million+",[107,156,157,160],{},[122,158,159],{},"Flagship model",[122,161,162],{},[17,163,164],{},"Google Gemini 3.1 Pro",[46,166,168],{"id":167},"why-google-gemini-31-pro-won-the-timeline","Why Google Gemini 3.1 Pro won the timeline",[13,170,171,172,175,176,179,180,183,184,187,188,99],{},"Gemini 3.1 Pro joined GenAI.mil ",[17,173,174],{},"8 weeks after its commercial launch"," in February 2026. According to the ",[17,177,178],{},"Pentagon AI Chief",", the goal is explicit: ",[17,181,182],{},"avoid single-vendor dependence",". The line said publicly — \"relying on a single model is never a good thing\" — is a thinly veiled critique of the exclusive contract the DoD had previously begun to concentrate on ",[17,185,186],{},"Anthropic",", before the ",[83,189,191],{"href":190},"\u002Fnews\u002Ftrump-anthropic-reconciliation-pentagon-dario-amodei-mythos-april-2026","Trump-Anthropic crisis and Pentagon-Dario Amodei reconciliation",[13,193,194,195,198],{},"In parallel, Google secured a contract extension to deploy Gemini on ",[17,196,197],{},"classified networks",", broadening scope to \"any lawful government purpose.\" That's a major shift — Google had, for years, refused certain DoD contracts under internal pressure (Project Maven episode).",[46,200,202],{"id":201},"why-103000-agents-isnt-a-fluke","Why 103,000 agents isn't a fluke",[13,204,205],{},"At first glance the number looks inflated: private-sector averages for internal agentic platforms cap at a few thousand agents per year, not 100,000 in five weeks. Three reasons explain the DoD surge:",[207,208,210],"h3",{"id":209},"_1-a-massive-and-heterogeneous-user-base","1. A massive and heterogeneous user base",[13,212,213,214,217,218,99],{},"The DoD has ",[17,215,216],{},"2.9 million people"," between active military, reservists and civilians. Each branch, each base, each squadron has specific workflows. A single agent only covers a very local use case — so you have to ",[17,219,220],{},"multiply agents by site",[207,222,224],{"id":223},"_2-an-already-secured-framework","2. An already secured framework",[13,226,227,228,231],{},"Users don't have to wonder if the tool is compliant: GenAI.mil is ",[17,229,230],{},"already accredited"," at the right security levels. That eliminates the main friction to AI experimentation in companies (the fear of violating compliance).",[207,233,235],{"id":234},"_3-a-culture-of-operational-efficiency","3. A culture of operational efficiency",[13,237,238,239,242],{},"Military personnel are trained to ",[17,240,241],{},"automate, standardize, document",". The match with a low-code agent generation platform is almost perfect. Where a civilian SMB hesitates, a sergeant immediately creates the agent that kills their weekly tedium.",[46,244,246],{"id":245},"implications-for-the-private-sector","Implications for the private sector",[13,248,249],{},"This adoption has three major consequences for software vendors and AI startups:",[207,251,253],{"id":252},"a-low-code-agents-are-no-longer-a-gadget","A. Low-code agents are no longer a gadget",[13,255,256,257,261,262,265],{},"When ",[83,258,260],{"href":259},"\u002Fnews\u002Fanthropic-native-app-builder-claude-lovable-competitor-vibe-coding-leak","Pentagon, Anthropic, Cursor and Lovable"," all tell the same story, you're entered the phase where ",[17,263,264],{},"any knowledge worker can produce their own agent",". The role of classic SaaS vendors shrinks; the role of broker platforms (orchestration, security, monitoring) explodes.",[207,267,269],{"id":268},"b-defense-becomes-a-tier-1-market-for-ai","B. Defense becomes a tier 1 market for AI",[13,271,272,273,28,277,280],{},"With ",[83,274,276],{"href":275},"\u002Fnews\u002Fshield-ai-2-billion-series-g-aechelon-hivemind-defense-ai-april-2026","Shield AI at $2B valuation",[17,278,279],{},"Scout AI"," just raising $100M Series A for unmanned warfare AI models, we're seeing a convergence: DoD budgets are shifting from hardware to software-as-a-service. The opportunity is massive — and politically sensitive.",[207,282,284],{"id":283},"c-google-takes-back-the-lead-partially","C. Google takes back the lead, partially",[13,286,287,288,292,293,297,298,99],{},"For 18 months, Google appeared to lose AI leadership to OpenAI and Anthropic. The expanded GenAI.mil contract, combined with the ",[83,289,291],{"href":290},"\u002Fnews\u002Fgoogle-anthropic-40-billion-investment-two-phase-cloud-tpu-april-2026","$40B investment in Anthropic"," and the ",[83,294,296],{"href":295},"\u002Fnews\u002Fgoogle-cloud-next-2026-gemini-enterprise-agent-platform-a2a-tpu8-april-2026","strong return of Gemini Enterprise at Cloud Next 2026",", traces a coherent strategy: Google is the ",[17,299,300],{},"plumber of enterprise agentic AI",[46,302,304],{"id":303},"open-questions","Open questions",[13,306,307],{},"The raw 103,000 agent count masks open questions:",[309,310,311,322,328,334],"ol",{},[312,313,314,317,318,321],"li",{},[17,315,316],{},"How many agents are actually used regularly?"," Many may have been created in test then abandoned. The key metric is ",[17,319,320],{},"recurrent agent runs",", not the number of agents built.",[312,323,324,327],{},[17,325,326],{},"What's the error rate?"," An agent hallucinating on a routine brief is annoying. An agent hallucinating on operational targeting is catastrophic. The DoD has not published reliability metrics.",[312,329,330,333],{},[17,331,332],{},"Ethical implications."," Vibe-coding an agent that summarizes patrol reports is benign. Vibe-coding an agent that prioritizes targets or interprets drone imagery is much less so. The line is fuzzy.",[312,335,336,339],{},[17,337,338],{},"Reverse vendor lock-in."," By diversifying models (Anthropic, OpenAI, Meta, Google), the DoD reduces vendor lock-in — but it creates lock-in on its own broker layer, which becomes critical infrastructure.",[46,341,343],{"id":342},"conclusion-the-worlds-largest-agentic-experiment-runs-at-the-dod","Conclusion: the world's largest agentic experiment runs at the DoD",[13,345,346,347,350],{},"No private company has managed to mobilize 1 million users and 100,000 agents in five weeks. The Pentagon just did. Whether you find that inspiring or alarming, this is the ",[17,348,349],{},"world's largest agentic experiment",", and its outcome will determine how low-code agents are adopted (or regulated) across the rest of society.",[13,352,353,354,358,359,363],{},"For developers and founders who want to ride this wave, our guide ",[83,355,357],{"href":356},"\u002Fblog\u002Fcreate-app-24h-idea-to-deployment","build an app in 24h, from idea to deployment"," is the simplest entry point, and our analysis ",[83,360,362],{"href":361},"\u002Fblog\u002Fvibecoding-for-non-technical-founders","vibe coding for non-technical founders"," shows how non-coders are now building as fast as coders.",[365,366],"hr",{},[13,368,369],{},[370,371,372,373,376,377,99],"em",{},"The DoD's massive agentic adoption proves a simple thesis: when you give non-developers a tool to automate their work, they build ",[17,374,375],{},"more"," of them than you thought possible. The same phenomenon is happening in finance with ",[83,378,380],{"href":379},"\u002Fnews\u002Frogo-160-million-series-d-kleiner-perkins-felix-agent-finance-april-2026","Rogo and its agent Felix that just raised $160M",{"title":382,"searchDepth":383,"depth":383,"links":384},"",2,[385,386,387,388,394,399,400],{"id":48,"depth":383,"text":49},{"id":77,"depth":383,"text":78},{"id":167,"depth":383,"text":168},{"id":201,"depth":383,"text":202,"children":389},[390,392,393],{"id":209,"depth":391,"text":210},3,{"id":223,"depth":391,"text":224},{"id":234,"depth":391,"text":235},{"id":245,"depth":383,"text":246,"children":395},[396,397,398],{"id":252,"depth":391,"text":253},{"id":268,"depth":391,"text":269},{"id":283,"depth":391,"text":284},{"id":303,"depth":383,"text":304},{"id":342,"depth":383,"text":343},"ai","2026-04-30T10:00:00","Over 103,000 AI agents built and 1.1 million sessions logged on GenAI.mil since Agent Designer launched. The Pentagon is industrializing vibe coding with Google Gemini 3.1 Pro. The U.S. Department of Defense becomes one of the largest low-code agent platforms in the world.","md","\u002Fpentagone-100000-agents-ia-vibe-coding-genai-mil-2026.jpg",{},true,"\u002Fnews\u002Fpentagon-100000-ai-agents-vibe-coding-genai-mil-google-gemini-april-2026",{"title":6,"description":403},"103,000 AI agents, 1.1M sessions on GenAI.mil. The DoD massively adopts Google Gemini 3.1 Pro and Agent Designer to vibe-code its workflows.","Pentagon: 100,000 AI agents vibe-coded in 5 weeks on GenAI.mil","news\u002Fpentagon-100000-ai-agents-vibe-coding-genai-mil-google-gemini-april-2026",[414,415,416,417,418,419,420,421],"pentagon","dod","genai-mil","vibe-coding","google-gemini","agent-designer","low-code","defense-ai","pentagon-100000-ai-agents-vibe-coding-genai-mil-google-gemini-april-2026","odcRtAC7gNN8419vIwmWGPoGD67fg0GWUbcat-eEU3Y",[425,432,438],{"title":426,"description":427,"path":428,"date":429,"image":430,"category":431},"Cursor cloud agents now build, host and publish live apps","Cursor's cloud agents can scaffold a web app, provision Postgres and Cursor SSO, run security checks, then publish a shareable URL — without a separate hosting account.","\u002Fnews\u002Fcursor-live-apps-cloud-agents-deploy-postgres-auth","2026-11-07T08:00:00","\u002Fimages\u002Fnews\u002Fcursor-live-apps-cloud-agents-deploy-postgres-auth.jpg","vibecoding",{"title":433,"description":434,"path":435,"date":436,"image":437,"category":401},"Google makes Gemini Omni 1.2 Flash generally available with 15-second video","Google DeepMind cut Omni Flash pricing by 30% to $0.07 per second at 720p, raised clip length from 10 to 15 seconds, and opened the model on the Gemini API and Google Cloud.","\u002Fnews\u002Fgoogle-gemini-omni-1-2-flash-15-second-video","2026-11-06T08:00:00","\u002Fimages\u002Fnews\u002Fgoogle-gemini-omni-1-2-flash-15-second-video.jpg",{"title":439,"description":440,"path":441,"date":442,"image":443,"category":401},"Anthropic reportedly signs its first publisher licensing deal, with the AP","Adweek reports that Anthropic struck a content and data licensing deal with the Associated Press just before the US midterms, its first with a digital publisher. Neither company has confirmed it on the record, and the terms are undisclosed.","\u002Fnews\u002Fanthropic-first-publisher-licensing-deal-associated-press","2026-11-05T12:35:00","\u002Fimages\u002Fnews\u002Fanthropic-ap-publisher-licensing-deal.jpg",{"en":445,"fr":678,"es":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":446,"category":401,"date":402,"description":403,"extension":404,"image":405,"meta":675,"navigation":407,"path":408,"seo":676,"seoDescription":410,"seoTitle":411,"stem":412,"tags":677,"translationKey":422,"updatedAt":8,"__hash__":423},{"type":10,"value":447,"toc":658},[448,464,466,474,480,482,486,492,540,542,554,558,560,562,564,570,572,576,578,582,584,586,588,594,596,602,604,612,614,616,636,638,642,648,650],[13,449,15,450,20,452,24,454,28,456,32,458,36,460,40,462,44],{},[17,451,19],{},[17,453,23],{},[17,455,27],{},[17,457,31],{},[17,459,35],{},[17,461,39],{},[17,463,43],{},[46,465,49],{"id":48},[13,467,52,468,56,470,28,472,63],{},[17,469,55],{},[17,471,59],{},[17,473,62],{},[13,475,66,476,70,478,74],{},[17,477,69],{},[17,479,73],{},[46,481,78],{"id":77},[13,483,81,484,87],{},[83,485,86],{"href":85},[13,487,90,488,95,490,99],{},[83,489,94],{"href":93},[17,491,98],{},[101,493,494,502],{},[104,495,496],{},[107,497,498,500],{},[110,499,112],{},[110,501,115],{},[117,503,504,510,516,524,532],{},[107,505,506,508],{},[122,507,124],{},[122,509,55],{},[107,511,512,514],{},[122,513,131],{},[122,515,134],{},[107,517,518,520],{},[122,519,139],{},[122,521,522],{},[17,523,144],{},[107,525,526,528],{},[122,527,149],{},[122,529,530],{},[17,531,154],{},[107,533,534,536],{},[122,535,159],{},[122,537,538],{},[17,539,164],{},[46,541,168],{"id":167},[13,543,171,544,175,546,179,548,183,550,187,552,99],{},[17,545,174],{},[17,547,178],{},[17,549,182],{},[17,551,186],{},[83,553,191],{"href":190},[13,555,194,556,198],{},[17,557,197],{},[46,559,202],{"id":201},[13,561,205],{},[207,563,210],{"id":209},[13,565,213,566,217,568,99],{},[17,567,216],{},[17,569,220],{},[207,571,224],{"id":223},[13,573,227,574,231],{},[17,575,230],{},[207,577,235],{"id":234},[13,579,238,580,242],{},[17,581,241],{},[46,583,246],{"id":245},[13,585,249],{},[207,587,253],{"id":252},[13,589,256,590,261,592,265],{},[83,591,260],{"href":259},[17,593,264],{},[207,595,269],{"id":268},[13,597,272,598,28,600,280],{},[83,599,276],{"href":275},[17,601,279],{},[207,603,284],{"id":283},[13,605,287,606,292,608,297,610,99],{},[83,607,291],{"href":290},[83,609,296],{"href":295},[17,611,300],{},[46,613,304],{"id":303},[13,615,307],{},[309,617,618,624,628,632],{},[312,619,620,317,622,321],{},[17,621,316],{},[17,623,320],{},[312,625,626,327],{},[17,627,326],{},[312,629,630,333],{},[17,631,332],{},[312,633,634,339],{},[17,635,338],{},[46,637,343],{"id":342},[13,639,346,640,350],{},[17,641,349],{},[13,643,353,644,358,646,363],{},[83,645,357],{"href":356},[83,647,362],{"href":361},[365,649],{},[13,651,652],{},[370,653,372,654,376,656,99],{},[17,655,375],{},[83,657,380],{"href":379},{"title":382,"searchDepth":383,"depth":383,"links":659},[660,661,662,663,668,673,674],{"id":48,"depth":383,"text":49},{"id":77,"depth":383,"text":78},{"id":167,"depth":383,"text":168},{"id":201,"depth":383,"text":202,"children":664},[665,666,667],{"id":209,"depth":391,"text":210},{"id":223,"depth":391,"text":224},{"id":234,"depth":391,"text":235},{"id":245,"depth":383,"text":246,"children":669},[670,671,672],{"id":252,"depth":391,"text":253},{"id":268,"depth":391,"text":269},{"id":283,"depth":391,"text":284},{"id":303,"depth":383,"text":304},{"id":342,"depth":383,"text":343},{},{"title":6,"description":403},[414,415,416,417,418,419,420,421],{"id":679,"title":680,"author":7,"author_avatar":8,"body":681,"category":401,"date":402,"description":1044,"extension":404,"image":405,"meta":1045,"navigation":407,"path":1046,"seo":1047,"seoDescription":1048,"seoTitle":1049,"stem":1050,"tags":1051,"translationKey":422,"updatedAt":8,"__hash__":1053},"news_fr\u002Ffr\u002Fnews\u002Fpentagone-100000-agents-ia-vibe-coding-genai-mil-google-gemini-avril-2026.md","Pentagone : 100 000 agents IA vibe-codés en cinq semaines sur GenAI.mil avec Gemini 3.1 Pro",{"type":10,"value":682,"toc":1027},[683,710,714,728,738,742,750,761,819,823,844,851,855,858,862,872,876,883,887,894,898,901,905,917,921,931,935,951,955,958,988,992,999,1012,1014],[13,684,685,686,688,689,692,693,696,697,700,701,703,704,706,707,709],{},"Le 28 avril 2026, lors du ",[17,687,19],{},", le ",[17,690,691],{},"Pentagone"," a confirmé un chiffre que peu d'entreprises privées peuvent égaler : ",[17,694,695],{},"plus de 103 000 agents IA construits"," et ",[17,698,699],{},"1,1 million de sessions agent enregistrées"," sur ",[17,702,35],{},", sa plateforme générative interne — en moins de cinq semaines après l'arrivée d'",[17,705,73],{}," de Google. Officiers, analystes et civils du ",[17,708,43],{}," vibe-codent en masse pour automatiser leurs workflows, du tri de documents au reporting opérationnel.",[46,711,713],{"id":712},"genaimil-la-plateforme-qui-change-la-donne","GenAI.mil, la plateforme qui change la donne",[13,715,716,717,720,721,696,724,727],{},"Lancée le ",[17,718,719],{},"9 décembre 2025",", GenAI.mil a accumulé ",[17,722,723],{},"500 000 utilisateurs en une semaine",[17,725,726],{},"1 million d'utilisateurs en un mois",", sans latence, sans downtime. C'est l'une des adoptions enterprise les plus rapides jamais observées sur une plateforme IA, publique ou privée.",[13,729,730,731,734,735,737],{},"L'architecture est mixte : modèles d'",[17,732,733],{},"Anthropic, OpenAI, Meta et désormais Google Gemini 3.1 Pro",", intégrés derrière un broker maison. Les utilisateurs accèdent au tout via une interface unique. Et depuis l'ajout d'",[17,736,73],{}," au catalogue, le pic d'usage a basculé vers la création d'agents.",[46,739,741],{"id":740},"le-vibe-coding-entre-dans-le-dod","Le « vibe coding » entre dans le DoD",[13,743,744,745,749],{},"Le mécanisme est exactement celui que les développeurs grand public connaissent bien depuis ",[83,746,748],{"href":747},"\u002Ffr\u002Fblog\u002Fno-code-vs-vibecoding-vs-dev-classique-guide-startups","Lovable, Bolt et Cursor",". Un officier ouvre Agent Designer, décrit en langage naturel ce qu'il veut accomplir (« crée un agent qui compile chaque matin les rapports de patrouille des dernières 24h en un brief format SITREP »), et l'outil génère l'agent en autonomie.",[13,751,752,753,757,758,99],{},"Pas de syntaxe, pas de SDK à apprendre, pas de tickets soumis à une équipe IT centrale. C'est l'incarnation enterprise de ce que ",[83,754,756],{"href":755},"\u002Ffr\u002Fnews\u002Fcursor-3-interface-agents-parallele-vibe-coding-2026","Cursor 3 et son interface agents parallèle"," ont popularisé chez les développeurs civils — appliqué à un public de ",[17,759,760],{},"non-coders au sein du gouvernement",[101,762,763,773],{},[104,764,765],{},[107,766,767,770],{},[110,768,769],{},"Métrique GenAI.mil",[110,771,772],{},"Valeur",[117,774,775,782,790,800,810],{},[107,776,777,780],{},[122,778,779],{},"Lancement",[122,781,719],{},[107,783,784,787],{},[122,785,786],{},"Utilisateurs (J+30)",[122,788,789],{},"1 000 000+",[107,791,792,795],{},[122,793,794],{},"Agents construits",[122,796,797],{},[17,798,799],{},"103 000+",[107,801,802,805],{},[122,803,804],{},"Sessions agent",[122,806,807],{},[17,808,809],{},"1,1 million+",[107,811,812,815],{},[122,813,814],{},"Modèle phare",[122,816,817],{},[17,818,164],{},[46,820,822],{"id":821},"pourquoi-google-gemini-31-pro-a-remporté-la-timeline","Pourquoi Google Gemini 3.1 Pro a remporté la timeline",[13,824,825,826,829,830,832,833,836,837,839,840,99],{},"Gemini 3.1 Pro a été ajouté à GenAI.mil ",[17,827,828],{},"8 semaines après son arrivée commerciale"," en février 2026. Selon le ",[17,831,178],{},", l'objectif est explicite : ",[17,834,835],{},"éviter la dépendance à un fournisseur unique",". La phrase prononcée publiquement — « relying on a single model is never a good thing » — est une critique à peine voilée du contrat exclusif que le DoD avait précédemment commencé à concentrer sur ",[17,838,186],{},", avant la ",[83,841,843],{"href":842},"\u002Ffr\u002Fnews\u002Ftrump-anthropic-reconciliation-pentagone-dario-amodei-mythos-avril-2026","crise Trump-Anthropic et la réconciliation Pentagone-Dario Amodei",[13,845,846,847,850],{},"Google a en parallèle obtenu une extension de contrat pour déployer Gemini sur ",[17,848,849],{},"réseaux classifiés",", élargissant le périmètre à « tout objectif gouvernemental légitime ». C'est un changement majeur — Google avait, pendant des années, refusé certains contrats DoD sous pression interne (épisode Project Maven).",[46,852,854],{"id":853},"pourquoi-103-000-agents-ne-sont-pas-un-fluke","Pourquoi 103 000 agents ne sont pas un fluke",[13,856,857],{},"À première vue, le chiffre paraît gonflé : la moyenne secteur privé pour des plateformes agentiques internes plafonne à quelques milliers d'agents par an, pas 100 000 en cinq semaines. Trois raisons expliquent l'envolée du DoD :",[207,859,861],{"id":860},"_1-une-base-dutilisateurs-colossale-et-hétérogène","1. Une base d'utilisateurs colossale et hétérogène",[13,863,864,865,868,869,99],{},"Le DoD compte ",[17,866,867],{},"2,9 millions de personnes"," entre militaires actifs, réservistes et civils. Chaque branche, chaque base, chaque escadrille a des workflows spécifiques. Un seul agent ne couvre qu'un cas d'usage très local — donc il faut ",[17,870,871],{},"multiplier les agents par site",[207,873,875],{"id":874},"_2-un-cadre-déjà-sécurisé","2. Un cadre déjà sécurisé",[13,877,878,879,882],{},"Les utilisateurs n'ont pas à se demander si l'outil est conforme : GenAI.mil est ",[17,880,881],{},"déjà accrédité"," sur les bons niveaux de sécurité. Cela élimine le frein principal à l'expérimentation IA en entreprise (la peur de violer une compliance).",[207,884,886],{"id":885},"_3-une-culture-defficacité-opérationnelle","3. Une culture d'efficacité opérationnelle",[13,888,889,890,893],{},"Les militaires sont entraînés à ",[17,891,892],{},"automatiser, standardiser, documenter",". Le matching avec une plateforme low-code de génération d'agents est presque parfait. Là où une PME civile hésite, un sergent crée immédiatement l'agent qui élimine son tedium hebdomadaire.",[46,895,897],{"id":896},"les-implications-pour-le-secteur-privé","Les implications pour le secteur privé",[13,899,900],{},"Cette adoption a trois conséquences majeures pour les éditeurs de logiciels et les startups IA :",[207,902,904],{"id":903},"a-les-agents-low-code-ne-sont-plus-un-gadget","A. Les agents low-code ne sont plus un gadget",[13,906,907,908,912,913,916],{},"Tant que ",[83,909,911],{"href":910},"\u002Ffr\u002Fnews\u002Fanthropic-app-builder-natif-claude-concurrent-lovable-bolt-v0-vibe-coding","Pentagon, Anthropic, Cursor et Lovable"," racontent la même histoire, c'est qu'on est entré dans la phase où ",[17,914,915],{},"n'importe quel knowledge worker peut produire son propre agent",". Le rôle des éditeurs SaaS classiques se réduit ; le rôle des plateformes broker (qui orchestrent, sécurisent, monitor) explose.",[207,918,920],{"id":919},"b-la-défense-devient-un-marché-tier-1-pour-lia","B. La défense devient un marché tier 1 pour l'IA",[13,922,923,924,696,928,930],{},"Avec ",[83,925,927],{"href":926},"\u002Ffr\u002Fnews\u002Fshield-ai-2-milliards-serie-g-aechelon-hivemind-defense-ai-avril-2026","Shield AI à 2 Md$ valorisation",[17,929,279],{}," qui vient de lever 100 M$ en série A pour les modèles IA dédiés à la guerre sans pilote, on assiste à une convergence : les budgets DoD se déplacent du hardware vers le software-as-a-service. L'opportunité est massive — et politiquement sensible.",[207,932,934],{"id":933},"c-google-reprend-la-main-partiellement","C. Google reprend la main, partiellement",[13,936,937,938,942,943,947,948,99],{},"Pendant 18 mois, Google a paru décrocher du leadership IA face à OpenAI et Anthropic. Le contrat GenAI.mil élargi, combiné à l'investissement ",[83,939,941],{"href":940},"\u002Ffr\u002Fnews\u002Fgoogle-anthropic-40-milliards-investissement-deux-phases-cloud-tpu-avril-2026","40 Md$ dans Anthropic"," et au ",[83,944,946],{"href":945},"\u002Ffr\u002Fnews\u002Fgoogle-cloud-next-2026-gemini-enterprise-agent-platform-a2a-tpu8-avril-2026","retour en force de Gemini Enterprise au Cloud Next 2026",", trace une stratégie cohérente : Google est le ",[17,949,950],{},"plombier de l'agentique enterprise",[46,952,954],{"id":953},"les-zones-dombre","Les zones d'ombre",[13,956,957],{},"Le chiffre brut de 103 000 agents masque des questions ouvertes :",[309,959,960,970,976,982],{},[312,961,962,965,966,969],{},[17,963,964],{},"Combien d'agents sont vraiment utilisés régulièrement ?"," Beaucoup peuvent avoir été créés en test puis abandonnés. La métrique-clé, ce sont les ",[17,967,968],{},"agent runs récurrents",", pas le nombre d'agents construits.",[312,971,972,975],{},[17,973,974],{},"Quel est le taux d'erreur ?"," Un agent qui hallucine sur un brief de routine, c'est embêtant. Un agent qui hallucine sur un targeting opérationnel, c'est catastrophique. Le DoD n'a pas publié de métrique de fiabilité.",[312,977,978,981],{},[17,979,980],{},"Les implications éthiques."," Vibe-coder un agent qui résume des rapports de patrouille est anodin. Vibe-coder un agent qui priorise des cibles ou interprète des images de drone l'est beaucoup moins. La frontière est floue.",[312,983,984,987],{},[17,985,986],{},"Le vendor lock-in inverse."," En diversifiant ses modèles (Anthropic, OpenAI, Meta, Google), le DoD réduit le lock-in fournisseur — mais il crée un lock-in sur sa propre couche broker, qui devient infrastructure critique.",[46,989,991],{"id":990},"conclusion-la-plus-grande-expérimentation-agentique-du-monde-tourne-au-dod","Conclusion : la plus grande expérimentation agentique du monde tourne au DoD",[13,993,994,995,998],{},"Aucune entreprise privée n'a réussi à mobiliser 1 million d'utilisateurs et 100 000 agents en cinq semaines. Le Pentagone vient de le faire. Que vous trouviez ça inspirant ou inquiétant, c'est la ",[17,996,997],{},"plus grande expérimentation agentique au monde",", et son issue déterminera comment les agents low-code seront adoptés (ou régulés) dans le reste de la société.",[13,1000,1001,1002,1006,1007,1011],{},"Pour les développeurs et fondateurs qui veulent capter cette vague, notre guide ",[83,1003,1005],{"href":1004},"\u002Ffr\u002Fblog\u002Fcreer-app-24h-idee-deploiement","créer une app en 24h, de l'idée au déploiement"," reste le point d'entrée le plus simple, et notre analyse ",[83,1008,1010],{"href":1009},"\u002Ffr\u002Fblog\u002Fvibecoding-pour-founders-non-techniques","vibe coding pour founders non techniques"," montre comment les non-coders construisent maintenant aussi vite que les coders.",[365,1013],{},[13,1015,1016],{},[370,1017,1018,1019,1022,1023,99],{},"L'adoption massive de l'agentique au DoD prouve une thèse simple : quand vous donnez à des non-développeurs un outil pour automatiser leur travail, ils en construisent ",[17,1020,1021],{},"plus"," que ce que vous pensiez possible. Le même phénomène est en train de se produire dans la finance avec ",[83,1024,1026],{"href":1025},"\u002Ffr\u002Fnews\u002Frogo-160-millions-serie-d-kleiner-perkins-felix-agent-finance-avril-2026","Rogo et son agent Felix qui vient de lever 160 M$",{"title":382,"searchDepth":383,"depth":383,"links":1028},[1029,1030,1031,1032,1037,1042,1043],{"id":712,"depth":383,"text":713},{"id":740,"depth":383,"text":741},{"id":821,"depth":383,"text":822},{"id":853,"depth":383,"text":854,"children":1033},[1034,1035,1036],{"id":860,"depth":391,"text":861},{"id":874,"depth":391,"text":875},{"id":885,"depth":391,"text":886},{"id":896,"depth":383,"text":897,"children":1038},[1039,1040,1041],{"id":903,"depth":391,"text":904},{"id":919,"depth":391,"text":920},{"id":933,"depth":391,"text":934},{"id":953,"depth":383,"text":954},{"id":990,"depth":383,"text":991},"Plus de 103 000 agents IA bâtis et 1,1 million de sessions enregistrées sur GenAI.mil depuis l'arrivée d'Agent Designer. Le Pentagone industrialise le vibe coding avec Google Gemini 3.1 Pro. Le ministère de la Défense américain devient l'un des plus gros utilisateurs au monde de l'agentique low-code.",{},"\u002Ffr\u002Fnews\u002Fpentagone-100000-agents-ia-vibe-coding-genai-mil-google-gemini-avril-2026",{"title":680,"description":1044},"103 000 agents IA, 1,1M de sessions sur GenAI.mil. Le DoD adopte massivement Google Gemini 3.1 Pro et Agent Designer pour vibe-coder ses workflows.","Pentagone : 100 000 agents IA vibe-codés en 5 semaines sur GenAI.mil","fr\u002Fnews\u002Fpentagone-100000-agents-ia-vibe-coding-genai-mil-google-gemini-avril-2026",[1052,415,416,417,418,419,420,421],"pentagone","_wYpcARNaKB1HOhFC8jI2hXiP6JYH0ycgkEyYp_DR1A"]