Google just priced the acquihire of coding expertise at $1.5 billion — triple what Microsoft paid for GitHub, but this time they're not buying the platform, just the people who know how to make AI write code.
The Summary
- Google is negotiating a $1.5 billion+ talent deal with Mechanize, a San Francisco startup training AI agents to code, structured as an acquihire with non-exclusive tech licensing
- This marks Google's third major coding-focused acquisition workaround in two years, following similar deals for Windsurf and Character AI
- The structure dodges antitrust scrutiny while securing model evaluation and development talent for Google's Antigravity agentic coding platform
The Signal
Google is paying premium prices for something it already has armies of: engineers who can code. But these aren't just engineers. They're the people who've figured out how to teach machines to write software that works. That's a different skill entirely, and Google knows it.
Mechanize trains AI agents specifically for coding tasks, which has become the killer app of the current AI wave. Not chatbots. Not image generation. Code. The thing that actually builds products and generates revenue. When Microsoft bought GitHub for $7.5 billion in 2018, they were buying a platform. Google is paying $1.5 billion just for the knowledge of how to make AI do what GitHub's millions of developers do manually.
"The talent Google could acquire from Mechanize would work on model evaluation and development."
The deal structure tells you everything about the current moment in AI. Full acquisitions draw antitrust fire. So Google has perfected the workaround: hire the team, license the tech non-exclusively, call it a partnership. They did it with Windsurf last year (whose CEO Varun Mohan now runs Google's Antigravity coding platform). They did it with Character AI in 2024, bringing back cofounder Noam Shazeer. Now Mechanize.
This isn't about acquiring a product. It's about acquiring institutional knowledge that can't be rebuilt quickly:
- How to evaluate whether AI-generated code actually works at scale
- How to train models that understand not just syntax but software architecture
- How to build agents that can debug, refactor, and iterate without human handholding
That last piece matters most. Writing code is table stakes now. Cursor does it. GitHub Copilot does it. Google's own tools do it. But building an agent that can maintain a codebase, understand dependencies, and ship features end-to-end? That's still rare. Mechanize apparently cracked something Google wants.
The Implication
If Google is willing to spend this much on coding agent expertise, two things are true. First, the internal belief is that agentic coding will be a winner-take-most market, and they're behind. Second, the talent shortage in AI isn't about ML researchers anymore. It's about people who can make AI agents do useful work in production environments.
Watch for more of these hybrid deals. Big Tech has learned that paying $1.5 billion to skip two years of R&D is cheaper than falling behind in the agent race. For founders building in the agent space, the exit path is clear: get good enough that Google, Microsoft, or Anthropic would rather buy your team than compete with you. For everyone else, understand that the code you write today might be the last generation of purely human software engineering. The agents are learning fast.