Coinbase Runs 95% AI-Written Code While Keeping Humans in Charge

Coinbase just dropped a number that’s hard to ignore. Over 95% of its code is now written with AI assistance — a figure that would’ve seemed absurd at a major crypto exchange just a few years back. And the company isn’t treating it as some experimental side project. It’s core to how Coinbase builds software now.

Rob Witoff from Coinbase put it plainly: AI is crucial for execution tasks inside the company. That framing matters. Execution, not strategy. Routine, not judgment. Witoff’s point is basically that AI handles the heavy mechanical lifting — the repetitive, time-consuming code that developers used to grind through manually — while humans stay in the driver’s seat on anything that requires actual thinking about direction.

What 95% Actually Means on the Ground

Let’s be clear about what Coinbase is and isn’t saying. The 95% figure covers code written with AI assistance, not code written entirely without human involvement. There’s a difference. Developers aren’t sitting back watching a machine build the product. They’re reviewing, steering, correcting. But the raw volume of code being generated — the first drafts, the boilerplate, the repetitive structures — that’s almost entirely AI now.

The speed implications alone are significant. Traditional software development at a company Coinbase’s scale means hundreds of engineers writing, reviewing, and debugging millions of lines of code. Automating the bulk of initial code generation doesn’t just save time — it changes the rhythm of how a product gets built. Bugs that used to creep in through manual fatigue get caught earlier. Cycles that took weeks can compress.

Coinbase is pretty open about the tradeoff, though. AI-generated code still has to meet the company’s security standards and internal protocols. That’s not trivial for a crypto exchange sitting on billions in customer assets. Every line of code is potentially a vulnerability. So the efficiency gain from automation has to be weighed against the scrutiny those automated outputs still require from human engineers. The company acknowledged that ensuring AI-generated code aligns with their standards is a real challenge — not a solved problem.

Humans Aren’t Going Anywhere

Coinbase is pretty direct on one thing: high-agency humans are still essential. That’s the phrase the company uses — high-agency. It’s a deliberate choice of words. They’re not just saying humans are still around. They’re saying the humans who remain need to be capable of independent judgment, strategic thinking, and making calls that no model can make for them.

That’s probably the right framing for where AI tooling actually sits right now. Models can write solid code. They can’t decide what to build, why to build it, or when a product decision is bad for users even if it’s technically executable. Those calls stay with people.

Read also: Coinbase Smart Wallet Update Targets Multi-Chain dApp Confusion for Base Users

So what does the workforce look like when 95% of your code is AI-assisted? Coinbase didn’t get specific about headcount or hiring plans. No details on whether this shift has reduced engineering staff or changed how the company recruits. That’s a gap in what’s been shared publicly. What the company did say is that the human role has shifted toward oversight and direction — less typing, more deciding.

And that’s not unique to Coinbase. Across the broader tech and crypto industry, AI coding tools have moved fast from novelty to standard practice. The question most companies are still working through is how to structure teams around that shift. Coinbase seems to be further along that curve than most, at least by the numbers.

Competitive Edge or Necessary Bet

The crypto sector moves fast. Really fast. New protocols, new regulatory pressures, new competitors — the development pace that exchanges have to maintain is brutal. If AI tooling genuinely compresses development cycles and reduces errors, that’s not just an efficiency story. It’s a competitive one.

Coinbase’s bet is that combining AI-driven speed with human strategic oversight gives it an edge over competitors still running more traditional development pipelines. Whether that plays out depends on execution — specifically on whether the human oversight layer is actually catching what it needs to catch, and whether the AI-generated code holds up under real-world conditions at scale.

More context: Hyperliquid Joins Circle and Coinbase in a Deal That Squeezes USDCs Revenue Model

No timeline has been given for any further expansion of AI’s role. No specific metrics on error rates, deployment speed, or code quality improvements were shared publicly. Those would be the numbers that actually prove the thesis. For now, the 95% figure is the headline, and Witoff’s framing of AI as an execution tool — not a replacement for human judgment — is the company’s public position.

Coinbase’s security infrastructure has to hold regardless of who or what wrote the underlying code.