Google Gemini's Agent Designer lets Pentagon staff build AI agents in plain language, driving 103,000 tools and 1.1 million sessions in under five weeks.
Defense Department personnel have built more than 103,000 semi-autonomous artificial intelligence agents in under five weeks on a single internal platform. That pace surprised program officials and signals something more significant than a technology pilot.
GenAI.mil, the Pentagon's AI hub launched in December 2025 and powered by Google Gemini, has logged over 1.1 million agent sessions since going live, Yahoo News reported Thursday. Usage now averages around 180,000 sessions per week. Each session represents one agent invoked once by one user, meaning a widely adopted tool can accumulate thousands of sessions weekly while a low-traffic utility barely registers.
The creation engine is Gemini's Agent Designer, a no-code interface where users describe in plain language what they want an agent to do and the software generates it automatically. This approach, informally called vibe-coding, has collapsed the barrier between a capability request and a working tool. Military and civilian Pentagon staff who have never written a line of code can now build, deploy, and iterate on automated workflows themselves.
What agents actually handle
Paperwork is the dominant use case so far. Personnel have built agents that draft After Action Reports and formal staff estimates based on user inputs, tools that analyze imagery and produce written descriptions, and utilities that assist with financial data review and parsing of official strategy documents. All outputs are flagged as drafts for human review, not final submissions. Pentagon officials have been explicit about that design requirement.
The platform operates with an Authorization to Operate at Impact Level 5, the clearance tier for sensitive but unclassified data including law enforcement records and other controlled unclassified information. That designation matters: building agents on an already-cleared system, rather than waiting for each new tool to undergo its own approval process, is a core reason adoption has accelerated. Users inherit the platform's authorization, not an open-ended license.
Scale in context
These numbers land as the broader artificial intelligence industry is pressing hard into enterprise and government markets. Forbes noted in January that ChatGPT has 800 million monthly active users and $20 billion in annual recurring revenue, a scale that reflects how fast the consumer segment moved. Government adoption has historically lagged, held back by procurement requirements and security reviews. At 103,000 agents in five weeks, the Pentagon suggests that lag is compressing.
Google's main rival for enterprise AI is pursuing a parallel strategy. CNBC reported in March that OpenAI plans to nearly double its headcount to 8,000 by year-end, with most hires going into engineering, product, research, and what the company calls technical ambassadorship roles aimed at helping large organizations use its tools. OpenAI's CEO reportedly issued an internal code red in December after Gemini advances. Whatever the internal urgency, Google secured a significant foothold inside the Department of Defense before that hiring push takes full effect.
Consolidation is simultaneously reshaping the vendor landscape. Cohere and Aleph Alpha, two enterprise-focused AI startups, announced a merger this week backed by $600 million in new funding from Germany's Schwarz Group, SiliconAngle reported. That deal signals smaller players are betting that scale, not differentiation alone, determines long-term viability in large institutional markets.
The rapid proliferation of agents inside a classified-adjacent environment raises questions that adoption metrics do not answer. The EU's Artificial Intelligence Act has pushed risk tiering onto the agenda for regulated sectors globally. Whether U.S. defense procurement develops analogous review and oversight frameworks, or relies on existing security clearance structures, will matter when the first consequential agent failure occurs. At 103,000 tools and counting, that scenario is no longer hypothetical.
What comes next depends more on governance than on compute.
FAQ
What is GenAI.mil?
The Pentagon's internal AI platform, launched in December 2025 and built on Google Gemini. It gives military and civilian Defense Department personnel access to AI chat tools and the Agent Designer for building semi-autonomous agents.
What is an AI agent in this context?
A software tool that automates a specific task, such as generating draft documents or processing imagery. On GenAI.mil, agents are created through plain-language descriptions rather than code.
What data can these agents access?
GenAI.mil agents are cleared to Impact Level 5, covering sensitive but unclassified information including law enforcement and other controlled unclassified data. Classified systems remain separate.
What is vibe-coding?
Building software by describing desired behavior in natural language and letting an AI system generate the underlying logic automatically. No programming background is required from the user.