Google's AI adoption looks like everyone else's

Steve Yegge dropped a bomb on Twitter last week: Google's AI adoption curve looks identical to John Deere's. The tractor company.

His breakdown: 20% agentic power users, 20% outright refusers, 60% still stuck in chat-based tools like Cursor. Nothing that suggests the company building the models has any edge using them internally.

The kicker was his explanation. An 18-month hiring freeze means no fresh blood. No one walking in from the outside to point out how far Google has fallen behind. Just engineers marinating in their own mediocrity.

The rebuttal came fast

Addy Osmani, a Google engineering director, fired back within hours. Over 40,000 Google engineers use agentic coding weekly. They have access to custom models, orchestrators, agent loops, virtual SWE teams.

Demis Hassabis weighed in too, though his tweet got cut off in the thread. The message was clear: we're not John Deere.

Both stories are true

40,000 weekly users sounds impressive until you remember Google has roughly 180,000 employees. Even if every one of those 40K is a software engineer, that's still a minority. Yegge's 20% power user estimate fits.

The 60% stuck in chat tools? Also tracks. Having access to agentic systems doesn't mean engineers trust them enough to hand over actual work. Cursor is safe. Cursor is controllable. Cursor doesn't rewrite your entire codebase while you grab coffee.

The real tell is that this argument happened publicly at all. Companies confident in their internal AI adoption don't need executives defending it on Twitter.

The hiring freeze matters more than the tools

Yegge's point about the freeze is the one that sticks. When no one leaves and no one joins, institutional knowledge calcifies. The engineers who figured out agentic workflows 18 months ago are still there. So are the ones who decided AI was hype and went back to their terminals.

No external pressure. No new hires asking why the team isn't using the tools everyone else has adopted. Just inertia.

Google built the models. They have the infrastructure. They have teams running sophisticated agentic systems. But having the tools and changing how 100,000+ engineers work are different problems. One is technical. The other is organizational.

Google is just as bad at the second one as John Deere.