Google's internal programming agent Agent Smith has become so popular that it's been throttled, allowing employees to use their phones to command the backend to write code.

According to monitoring by 1M AI News, Google employees are using an internal AI agent called Agent Smith, which can automatically complete programming and other tasks. The tool has been forced to limit access due to a surge in users. The name likely pays homage to the antagonist Agent Smith from “The Matrix.”

Agent Smith is built on Google’s existing agent programming platform Antigravity and can access various internal systems, having been launched earlier this year. Compared to previous AI coding assistants, Agent Smith can more autonomously plan and execute complete workflows, and it supports asynchronous operation, allowing it to perform tasks independently in the background. Employees do not need to continuously operate their computers and can check progress and issue commands via their phones at any time, and it can also be used directly from Google’s internal chat platform. By connecting to the employee data system, Agent Smith can automatically retrieve relevant documents, eliminating the need for manual searches.

Google co-founder Sergey Brin mentioned in an all-hands meeting for the sales department in early March that AI agents will become an important direction for Google this year, and hinted that the company is developing a tool similar to OpenClaw (it is still uncertain whether this is indeed Agent Smith). Google’s business head Philipp Schindler joked at the meeting that he can tell when Brin’s messages are sent by an agent.

Google is accelerating the internal adoption of AI tools across the board. Some employees have been informed that AI usage will be included in performance evaluations. The infrastructure department is also advancing an internal project called Project EAT, aimed at improving the adoption rate and standardization of AI tools. A Google spokesperson responded, “We are continually exploring new ways to build agents that can solve real problems, but we don’t have anything more to share at this time.”