Coinbase Says AI Now Writes Over 95% of Its Code

Coinbase says it has moved far beyond early experiments with artificial intelligence, with executives describing AI as now being deeply embedded in how the exchange builds and tests software. In the wake of a May workforce reduction, the company’s leaders pointed to AI as a central reason it has been able to speed up development and reorganize teams.

According to Coinbase executives, more than 95% of the company’s code is now written by or with large language models (LLMs). That figure marks a sharp jump from an earlier estimate Coinbase shared in February, when it said AI supported about 40% of its code.

Key takeaways

• Coinbase reports that between 95% and 100% of its code is written by or with LLMs, according to platform chief Rob Witoff.
• The company links its May layoffs to a need to restore the “speed and focus” of its startup-era operations, placing AI at the core.
• Witoff describes a “wide spectrum” of AI usage—from human-led work on core cryptography to fully automated prototyping.
• Coinbase says smaller, senior teams can handle work that previously required far larger groups, with AI agents running continuously.

From “helping” to “writing”: Coinbase’s evolving AI role

Coinbase cut roughly 14% of its workforce in May, laying off about 700 employees. In an email to staff, CEO Brian Armstrong said the company needed to “return to the speed and focus of our startup founding, with AI at our core,” framing AI as a major shift in how work moves.

Speaking to Cointelegraph, Coinbase’s head of platform, Rob Witoff, said the company has reached a state where AI is used on an everyday basis. “Effectively, 100% of our employees are using AI on a daily basis,” Witoff said. He added that “close to 100% of our code” is written by or with LLMs—“probably somewhere between 95% and 100%.”

The size of that jump matters because it signals how quickly AI adoption has progressed within crypto infrastructure, not just as a productivity tool but as part of the coding workflow. Coinbase’s earlier February estimate—AI supporting around 40% of its code—suggests a rapid acceleration in how the exchange operationalizes AI in engineering.

Human oversight stays central for high-stakes cryptography

Even with such high automation claims, Witoff emphasized that Coinbase’s use of AI isn’t uniform across all parts of its stack. He described a “wide spectrum” in the way AI is deployed, depending on risk and complexity.

“For example, when we’re writing core cryptography, we have industry-leading cryptographers that are meticulously researching and reviewing one line at a time.”

In other areas, AI plays a different role. Witoff said AI is used heavily to test and verify that code behaves correctly and to help check for vulnerabilities. However, he portrayed this work as still requiring more manual oversight than the prototyping pipeline.

“We’re using AI quite a bit to test and make sure the code we’ve written is working the way it should, there’s no vulnerabilities, we’re verifying the math,” Witoff said, adding that this verification process remains more hands-on. By contrast, he said internal prototyping is now “effectively a 100% automated” workflow.

For builders and investors watching how AI reshapes crypto companies, this distinction is important: the move toward AI-driven development appears strongest in low-to-medium risk areas like prototyping and iteration, while core security work remains tightly supervised by specialized experts.

Smaller teams, more senior judgment—and the “agent” layer

Coinbase also linked its AI-driven coding changes to a reorganization of how teams are structured. Witoff said the company has been able to move toward smaller, more senior groups, where two or three employees can handle work that previously required 10 or more people.

He noted that the May layoffs disproportionately affected junior roles, describing “a lot of junior development roles” among those impacted. While engineering saw major changes, he also said cuts extended beyond developer functions, including positions in marketing, legal, customer support, and compliance.

“For those smaller teams to work, for people to have the taste, the judgment, I think a lot about people having the battle scars so they know how to point agents in the right direction.”

The “agent” concept is a key part of the operational picture. Witoff said most Coinbase engineers now work with five to 10 AI agents running at any given time, and that these agents collectively perform coding work equivalent to about 1,200 employees.

Looking further ahead, Witoff suggested that by 2030 AI agents could do the equivalent work of 100,000 employees. While such projections are inherently speculative, the company’s internal framing underscores a broader industry belief: once AI agents are integrated into development processes, the bottleneck shifts from generating code to steering systems responsibly and validating outcomes—especially where security and correctness are non-negotiable.

AI-led restructuring is spreading across crypto

Coinbase’s story fits into a wider pattern seen across the sector this year, where multiple crypto firms cited AI-driven efficiency as they rebalanced headcount.

In March, crypto exchange Crypto.com cut about 12% of staff, according to Cointelegraph reporting, affecting roles that “do not adapt in our new world.” Earlier, Block CEO Jack Dorsey said he was cutting about 40% of the company’s workforce, describing AI-enabled changes to how companies build and operate.

Other crypto organizations mentioned in the same broader wave include Kraken, Gemini, Messari, and Dune, each of which has been reported to have adjusted staffing in part as they increased AI use.

For market participants, this matters beyond workforce headlines. When exchanges and crypto services reduce layers of staffing while increasing reliance on AI-driven workflows, it can change how quickly they ship product, how they manage operational risk, and how they allocate resources between core engineering and support functions.

There is still an open question for readers: how sustainable and measurable these gains are over time, especially for the parts of the stack that require specialized security review. Coinbase’s next signal will likely be whether its AI-heavy approach continues to hold up under stress—whether in scaling performance, minimizing vulnerabilities, or maintaining reliability as token and user activity grows.