Loading article…
Coinbase now writes over 95% of its code with AI assistance, up from 40% in February, reshaping its engineering teams after a 14% staff cut.
Over 95% of Coinbase’s code is now generated with AI assistance, a jump from the 40% figure reported in February, and the shift follows a 14% workforce reduction that cut 700 jobs in May 2024 [1]. The move signals a fundamental redesign of the exchange’s development process, with senior engineers overseeing AI‑produced drafts while human judgment remains essential for security‑critical work.
| At a glance | |
|---|---|
| AI‑written code | >95% |
| Prior AI usage | 40% (Feb 2024) |
| Workforce cut | 14% (700 staff) |
| AI agents output | ≈1,200 employee equivalents |
Rob Witoff, Coinbase’s head of platform, said that “close to 100% of our code, probably somewhere between 95% and 100%, is written by or with LLMs today” and that most engineers run five to ten AI agents simultaneously [1]. The AI‑generated drafts cover routine boilerplate and prototype development, while core cryptographic code still undergoes line‑by‑line human review. Witoff estimates the collective output of these agents matches the work of roughly 1,200 employees, and projects that by 2030 AI could emulate the effort of 100,000 staff [1].
The AI surge has allowed Coinbase to restructure around smaller, senior‑heavy teams. Tasks that once required ten or more developers can now be handled by two or three engineers, a change that accelerated the May layoffs, especially among junior developers [1]. Despite the reduction, the company stresses that “high‑agency humans” remain indispensable for strategic decisions and security oversight, underscoring a division of labor where AI handles execution and humans retain judgment [2].
Coinbase’s AI spend has stayed “flat” even as the volume of AI‑produced code has more than doubled, suggesting the firm is leveraging existing infrastructure rather than scaling budgets [1]. However, the exchange acknowledges that every line of AI‑generated code must still meet its stringent security standards, a non‑trivial requirement given the billions of customer assets it safeguards [2].
Coinbase’s rapid AI integration reshapes its development tempo and cost structure, but the ultimate test will be whether human oversight can keep pace with the speed and volume of AI‑produced code while maintaining the security standards demanded by a crypto exchange handling billions in assets.
Coverage is mostly measured — 114 of 125 reports stay neutral.
Every Monday — the token unlocks, Fed dates & catalysts set to move crypto and markets this week. So you’re never blindsided.
Free · 3-min read · one-click unsubscribe
AI-assisted synthesis by the TrendWatcher Editorial Desk · sourced from 3 outlets · Jul 16, 2026 · How we report
Pollak said his bet on on‑chain social applications and creator coins failed to drive crypto adoption, prompting a strategic pivot toward trading, payments, tokenization and AI.
Jordan Fish, also known on X as "Cobie," will oversee the Base app development, working to expand its capabilities beyond the Base ecosystem.
According to Coinbase’s head of platform, Rob Witoff, between 95% and 100% of the company’s code is now written with or by large language models.
AI integration enabled the company to reorganize into smaller, senior‑focused teams, with two or three engineers now able to handle work that previously required ten or more staff.
Coinbase reported that its AI spend has remained flat despite growing token usage.