Google is in active talks with San Francisco startup Mechanize on a deal worth over $1.5 billion, according to four people familiar with the conversations. The structure follows a pattern Google has used before: bring in the talent through a hire and secure the technology through a non-exclusive licensing agreement, rather than a full acquisition.
The Mechanize employees Google would absorb are expected to work on model evaluation and development. Both Google and Mechanize declined to comment.
Why Mechanize
Mechanize trains AI agents to write code. That puts it squarely in the most competitive corner of the current AI market. OpenAI’s Codex and Anthropic’s Claude Code have been pulling developer customers, and Google has reportedly struggled to keep pace with its own coding model efforts.
Mechanize raised $9.1 million earlier this year at a $500 million valuation. Its backers include former GitHub CEO Nat Friedman, Stripe CEO Patrick Collison, and podcaster Dwarkesh Patel. CEO Tamay Besiroglu previously cofounded Epoch AI, which focused on AI model evaluation.
Google’s acquihire playbook
This is not the first time Google has structured a deal this way to sidestep full-acquisition antitrust scrutiny. Last year it acquired Windsurf’s talent and licensed the technology after OpenAI attempted to buy the company outright. Windsurf’s CEO Varun Mohan now runs Google’s agentic coding platform, Antigravity. In 2024, Google rehired Character AI cofounder Noam Shazeer and paid for non-exclusive rights to Character AI’s technology. Shazeer has since left Google for OpenAI.
The operator angle
If the deal closes, Google gets dedicated model evaluation talent and a non-exclusive license to Mechanize’s training infrastructure for coding agents. For developers betting on which coding assistant improves fastest over the next 12 months, this is a signal worth tracking. Mechanize’s stated long-term goal, per its website, is “the full automation of valuable work across the economy.” For now, it’s focused on software engineering.
