Coinbase says 95% of its code now AI‑assisted

Over 95% of Coinbase’s software is now generated with AI assistance, a milestone the exchange says reshapes its development rhythm while keeping senior engineers in charge of strategic decisions【1】. The shift matters because it could accelerate product rollout and reduce human error in a sector where security breaches can cost billions.

AI as the execution engine

Coinbase frames AI as a tool for “execution tasks” – the repetitive, boiler‑plate coding that engineers previously wrote manually. According to Rob Witoff, the AI handles the bulk of first‑draft code, while developers review, steer, and correct the output. This division aims to compress development cycles that previously took weeks, and to catch fatigue‑related bugs earlier. The company stresses that every line of AI‑generated code still undergoes the same security reviews required for any production code, reflecting the exchange’s need to protect billions in customer assets【1】.

Human oversight remains essential

Despite the high AI contribution, Coinbase insists that “high‑agency humans” are still required for any decision that involves judgment, strategy, or product direction. The firm does not disclose whether the AI shift has altered headcount or hiring plans, noting only that engineers now focus more on oversight than on typing【1】. This mirrors a broader industry trend where AI coding tools have moved from novelty to standard practice, but the exact impact on team structures remains unclear.

Competitive edge or necessary evolution?

In a fast‑moving crypto market, the ability to ship updates quickly can be a competitive advantage. Coinbase’s claim is that combining AI‑driven speed with human strategic oversight could give it an edge over rivals still relying on traditional development pipelines. However, the exchange has not released concrete metrics on error rates, deployment speed, or code‑quality improvements, leaving the actual performance benefit unverified【1】.

What to watch

• AI‑generated code audit outcomes – any public disclosure of error‑rate or security‑incident data could confirm the efficiency claim.
• Headcount signals – hiring announcements or layoffs in Coinbase’s engineering teams may indicate how the AI shift affects staffing.
• Regulatory scrutiny – statements from regulators about software security standards for crypto exchanges could impact the viability of AI‑heavy development.

The 95% figure shows how far AI tooling has penetrated a major crypto exchange’s codebase, but the real test will be whether human oversight can consistently catch the vulnerabilities that AI alone might miss.