Coinbase AI Code Development Jumps From 40% to 95% in Months

Coinbase has reached a threshold that most software companies only talk about. Coinbase AI code development has reached a point where between 95% and 100% of the company’s code is now written by or with the direct assistance of artificial intelligence — a leap from roughly 40% in February. That’s not an incremental improvement. That’s a near-total transformation of how one of the world’s largest crypto platforms builds its products.

Key takeaways

• Between 95% and 100% of Coinbase’s code is now AI-written or AI-assisted, up from about 40% in February.
• Nearly all Coinbase employees use AI tools daily, with engineers running 5 to 10 AI agents simultaneously for coding tasks.
• The cumulative AI coding effort is equivalent to the output of roughly 1,200 human employees.
• Coinbase projects AI agents could handle work equivalent to 100,000 employees by 2030.
• These figures were shared by Rob Witoff, Coinbase’s Head of Platform, in a Cointelegraph interview.

Coinbase’s Extensive Use of AI in Code Development

The numbers Rob Witoff shared in a recent Cointelegraph interview are striking not just for their scale, but for their speed. Witoff, who serves as Coinbase’s Head of Platform, described a company that has moved faster on AI integration than almost any other name in fintech. From a 40% AI-assisted codebase in February to effectively 95-100% now — in a matter of months — is a pace that few engineering organizations have matched publicly.

Daily AI Adoption Across the Organization

This isn’t confined to a specialized AI team or a pilot program. According to Witoff, nearly all Coinbase employees use AI daily in their work. That kind of organization-wide adoption signals something deeper than tooling — it reflects a cultural shift in how the company thinks about labor, productivity, and engineering itself.

For context, daily AI use across an entire workforce is still rare. Most large companies report pockets of adoption, not universal integration. Coinbase appears to have compressed what typically takes years into a single product cycle.

AI Agents Running the Engineering Floor

On the engineering side, the mechanics are equally telling. Coinbase engineers typically run between 5 and 10 AI agents simultaneously while working on coding tasks. These aren’t passive autocomplete tools — they’re active agents handling work in parallel, allowing engineers to oversee multiple streams of development at once.

That multiplier effect is what makes the productivity claims plausible. When a single engineer manages a fleet of AI agents, the human-to-output ratio shifts dramatically. It’s less about replacing engineers and more about changing what a single engineer’s working day looks like.

Impact and Scale of AI Coding Work at Coinbase

Witoff put a concrete number on what all that AI activity adds up to: the combined coding work performed by AI at Coinbase is equivalent to roughly 1,200 human employees. For a company that has publicly navigated cost pressures and workforce decisions over the past few years, that figure carries real strategic weight.

It also reframes the standard narrative around AI and jobs. The question isn’t simply whether AI displaces workers — at Coinbase, the framing is about what AI enables the existing workforce to achieve. Whether that framing holds up under scrutiny is a different conversation, but the productivity math being cited is hard to ignore.

The 2030 Projection and What It Implies

If the current trajectory holds, Coinbase’s ambitions for AI-assisted development scale into genuinely uncharted territory. Witoff suggested the company expects AI capabilities to expand to the point where agents could handle work equivalent to 100,000 employees by 2030 — an eighty-fold increase from the current 1,200-employee equivalent.

Projections that far out carry inherent uncertainty, and it’s worth treating the 2030 figure as a directional signal rather than a firm forecast. But even as a signal, it tells you something important about how Coinbase’s leadership views the role of AI: not as a supplement to human engineering, but as the primary engine of it.

That kind of bet, made publicly and with specifics attached, puts pressure on competitors across the crypto and broader fintech space to articulate their own AI strategies with similar clarity. Coinbase isn’t just adopting AI — it’s defining what aggressive adoption looks like at scale.

What This Means for Fintech Engineering

The broader implication of Coinbase’s AI push is less about one company’s efficiency gains and more about what it signals for software engineering in financial technology overall. When a platform operating under heavy regulatory scrutiny — handling real user funds, real transactions, and real security requirements — moves to near-total AI-assisted code development, it sets a precedent the rest of the industry will have to respond to.

Speed, cost, and scale of output are changing simultaneously. The companies that treat AI integration as an optional upgrade risk falling behind on all three dimensions at once. Coinbase, through Witoff’s candid interview with Cointelegraph, has essentially drawn a new baseline for what AI-native development looks like in crypto.

FAQ

How much of Coinbase’s code is currently assisted or written by AI?

Between 95% and 100% of Coinbase’s code is now written by or with the assistance of AI, up from approximately 40% in February.

Do Coinbase employees use AI tools in their daily work?

Yes. According to Rob Witoff, Coinbase’s Head of Platform, nearly all employees at the company use AI daily as part of their work.

How many AI agents do engineers at Coinbase typically use when coding?

Coinbase engineers typically run between 5 and 10 AI agents simultaneously while working on coding tasks.

What are Coinbase’s expectations for AI coding capabilities by 2030?

Coinbase expects AI agents to be capable of handling work equivalent to 100,000 employees by 2030, compared to a current equivalent of roughly 1,200 employees.

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.