Coinbase AI Now Handles Work of 1,200 Developers, Witoff Says

Coinbase just crossed a threshold that most tech companies only talk about. Rob Witoff, the exchange’s platform head, said AI now writes between 95% and 100% of the company’s code — work that was previously handled by roughly 1,200 human employees.

That number is jarring on its own. But what makes it more striking is how fast it happened. Back in February, Witoff put AI’s share of coding at around 40%. So in a matter of months, Coinbase went from AI doing less than half the coding to AI doing basically all of it. That’s not a gradual shift. That’s a near-complete handover, and it happened fast enough that most people in the industry probably didn’t see it coming at that speed.

From 40% to Nearly Full Automation

The jump from 40% to somewhere between 95% and 100% is the kind of stat that sounds exaggerated until you look at how aggressively large tech companies have been pushing AI into their development pipelines. Coinbase isn’t alone in moving in this direction — plenty of firms have been quietly increasing AI’s role in writing, testing, and deploying code. But going public with numbers this high is pretty rare. Most companies stay vague about exactly how much of their software is machine-generated.

Witoff didn’t hedge much. He put the figure at 95% to 100%, and he tied it directly to workforce displacement — roughly 1,200 roles’ worth of output now handled by AI agents. That’s not a rounding error. That’s a large engineering team’s worth of labor, absorbed by automated systems.

The exact tools Coinbase uses for this haven’t been disclosed. No breakdown of which AI systems, which vendors, or how the workflow actually runs day-to-day. So there’s still a lot that’s unclear about the mechanics behind the numbers.

What It Means for Coinbase’s Operations

For Coinbase specifically, the efficiency case is obvious. Faster code iteration, faster deployment of features and updates, lower overhead on the development side. Witoff framed it around speed and precision — AI bringing both to tasks that used to require large teams of human developers coordinating across sprints and reviews.

And it’s not just about cutting costs, though that’s clearly part of it. The financial technology space moves fast. New products, regulatory changes, competitor moves — all of it demands software that can be updated and shipped quickly. If AI can compress that cycle dramatically, Coinbase probably sees it as a competitive edge, not just a cost-saving measure.

Read also: U.S. Government Moves Nearly 4,000 Bitcoin to Coinbase Prime in $250M Transfer

But the workforce side of this is harder to ignore. Twelve hundred employees’ worth of coding output is now automated. Whether those specific individuals lost jobs, moved into different roles, or were never hired in the first place — Coinbase hasn’t said. The company hasn’t disclosed detailed plans about what happens to workers whose previous functions are now handled by AI agents. That’s a gap in the story, and it’s a meaningful one.

There’s a broader conversation happening across the tech industry about what AI-driven automation actually does to employment. Some companies argue it frees up human workers for higher-level tasks. Others quietly reduce headcount. Coinbase’s position on that isn’t clear yet.

A Benchmark Other Companies Are Watching

What Coinbase is doing — or at least what Witoff is willing to say publicly — probably matters beyond the exchange itself. Other firms in fintech, crypto, and broader software development are watching how this plays out. If a major exchange can run on AI-generated code at this scale without obvious failures in product quality or security, it changes the calculus for everyone else.

It’s worth noting that Coinbase operates in a sector where code quality isn’t just a performance question — it’s a security question. Bugs in crypto infrastructure can mean lost funds, exploited contracts, or compromised user data. So the claim that AI can handle 95% to 100% of coding at this level carries more weight than it would at, say, a content platform. Either the AI tools are genuinely that good, or Coinbase has built serious human review layers around the output. Probably both, though details on that aren’t available.

Related: Circles $1.4 Billion Coinbase Bill Eats Half of USDC Revenue

Witoff’s disclosure is one of the more candid admissions from a major crypto company about how deeply AI has penetrated its core operations. Most executives in this space talk about AI as a tool that assists developers. Witoff is saying it pretty much replaced them — at least in terms of output volume.

Whether other exchanges or fintech firms follow with similar disclosures, or whether they quietly do the same thing without saying so, is an open question. Coinbase just made it harder to pretend the question doesn’t exist.

Witoff put the figure at 95% to 100%, tied to roughly 1,200 workers’ worth of output.