Coinbase says AI now writes 95%–100% of its code, up from 40% in February
Platform head Rob Witoff tied the shift to smaller senior teams after May’s 700-person layoff.
Coinbase head of platform Rob Witoff said the exchange now has “close to 100%” of its code written by or with large language models, putting the range at 95%–100%. He also said effectively all Coinbase employees use AI daily, framing the change as a companywide operating model shift after a May workforce reduction.
Key Takeaways
- Coinbase’s head of platform Rob Witoff put AI-assisted code generation at “somewhere between 95% and 100%” of the company’s codebase.
- Daily AI usage is effectively universal inside the company, with Witoff saying “Effectively, 100% of our employees are using AI on a daily basis here.”
- The new 95%–100% range is described as more than double Coinbase’s February estimate that 40% of its code was written with AI.
- Coinbase cut 14% of its workforce earlier in 2026, with 700 staff reductions in May.
Coinbase Puts AI-Written Code at 95%–100%, Up From 40% in February
Coinbase is now describing AI as the default path for shipping software, not a sidecar tool for a subset of engineers. Rob Witoff, the company’s head of platform, said “And close to 100% of our code, probably somewhere between 95% and 100%, is written by or with LLMs today.”
Witoff paired that with a broader adoption claim across the org: “Effectively, 100% of our employees are using AI on a daily basis here.” Taken together, the message is that AI is embedded across workflows, not confined to engineering.
The scale-up is also framed as rapid. The 95%–100% figure is described as more than double Coinbase’s February estimate that 40% of its code was written with AI, implying a sharp internal acceleration over a few months.
From 10-Person Pods to 2–3 Senior Engineers: The New Staffing Model
Management is explicitly tying the tooling shift to headcount efficiency. Witoff said the move enabled Coinbase to reorganize around smaller, more senior teams, with “two or three employees now capable of handling work that previously required 10 or more people.”
That posture lands in the context of workforce cuts. Coinbase cut 14% of its workforce earlier in 2026, and 700 staff were cut in May. CEO Brian Armstrong told employees in a May email that AI had “dramatically” changed the pace of work and Coinbase needed to “return to the speed and focus of our startup founding, with AI at our core.”
Witoff said the May layoffs hit junior roles disproportionately: “There were a lot of junior development roles that were impacted.” He added that cuts also extended across marketing, legal, customer support, and compliance, which matters for traders who map product velocity and regulatory capacity to execution risk.
Inside the Workflow: 5–10 Agents per Engineer and a 1,200 ‘Employee-Equivalent’ Claim
Coinbase is also leaning into agentic tooling as the unit of productivity. Witoff said most Coinbase engineers run five to 10 AI agents at any given time, and that these agents collectively perform coding work equivalent to about 1,200 employees.
He pushed the framing further out the curve, projecting that by 2030 Coinbase could see AI agents doing the equivalent work of 100,000 employees. Those are executive estimates, not audited operating metrics, but they set expectations for how aggressively Coinbase intends to compress labor inputs per unit of output.
Witoff also drew a clear line between speed and safety. He described a “wide spectrum” of AI reliance: core cryptography remains heavily human-driven, while prototyping is described as effectively automated. “For example, when we’re writing core cryptography, we have industry-leading cryptographers that are meticulously researching and reviewing one line at a time,” he said. He added: “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, but that’s a much more manual part than where we’re building internal prototypes, which is now effectively a 100% automated.”
Earnings-Season Proof Points for Traders Watching COIN’s Operating Leverage
The near-total AI-assisted coding claim is a cost-structure narrative that will be tested in disclosures, not soundbites. Coinbase also said its AI spend has remained “flat” despite growing token use, but no timeframe or dollar figures were provided.
For COIN watchers, the next earnings call is the first real checkpoint for quantified evidence behind “flat” AI spend, including any commentary on cloud and tooling line items, opex trajectory, or capex guidance. Hiring signals matter too: whether Coinbase continues reducing junior engineering roles post-May, or pivots toward more senior “agent-orchestration” profiles.
Operationally, the market will look for proof in product and reliability cadence. If Coinbase is generating 95%–100% of code with LLMs, the real-world tell should be faster shipping cycles without a corresponding rise in incidents. Another key signal is whether Coinbase replaces the 95%–100% range with a tighter internal metric or a clearer methodology for what counts as “written by or with” LLMs.
AI Productivity Claims Are a Cost-Structure Narrative—But the Metrics Need Verification
I treat Coinbase’s 95%–100% figure as a sentiment catalyst until the company shows repeatable, measurable outputs. The threshold that matters is whether this AI-heavy workflow produces visible operating leverage in reported expenses while maintaining reliability, especially given Witoff’s own distinction between security-critical cryptography and speed-oriented prototyping.
The real test is whether management can translate “flat” AI spend, smaller teams, and the 1,200 employee-equivalent framing into numbers that survive earnings scrutiny. If that holds, the setup starts to look structural rather than narrative-driven, and it matters because it would re-rate how traders model Coinbase’s cost base versus its ability to ship and compete.
