Pentagon workers have used a Google Gemini tool on the GenAI.mil platform to create approximately 100,000 AI 'agents' designed to handle data and automate online tasks on unclassified networks. The initiative marks a significant grassroots adoption of generative AI within the Defense Department, enabling personnel without formal coding expertise to develop custom automation solutions.
The move reflects a broader shift toward operationalizing AI at the tactical level, allowing individual units to streamline repetitive workflows rather than relying solely on top-down procurement. By empowering non-technical staff to 'vibe-code' agents, the Pentagon may accelerate internal efficiency gains across logistics, intelligence analysis, and administrative processes.
The use of unclassified networks suggests these agents are not yet integrated with sensitive or classified systems, potentially limiting their immediate operational impact. However, the rapid scaling—100,000 agents created—indicates strong demand and could pressure leadership to expand access to more secure environments.
Cost figures were not disclosed. The platform leverages existing Google infrastructure under the Defense Department's GenAI.mil program, which provides a sandboxed environment with guardrails to prevent data leakage or misuse. Training and compliance oversight remain critical as the number of user-generated agents grows.
Some security analysts caution that allowing broad creation of AI agents without centralized vetting could introduce vulnerabilities or inconsistent performance. The Pentagon has not yet announced formal review protocols for the agents, raising questions about long-term risk management in operational settings.