At OKX, the founder says artificial intelligence now writes 95% of the engineering pull requests. Star Xu disclosed the figure in a recent statement, offering a rare concrete look at just how deeply AI has burrowed into the daily work of a major crypto exchange.
The number covers pull requests — the code change requests developers submit for review. If accurate, it means the overwhelming majority of coding proposals at OKX are now generated by AI, not written by hand. That's a striking shift for any software company, let alone one handling billions in digital assets.
What 95% actually means inside OKX
An engineering pull request is the bread and butter of software development: a developer writes or modifies code, then asks colleagues to review and merge it. At a company of OKX's size, those requests pile up by the thousands. Xu's figure suggests AI tools are now the first author on nearly all of them.
That doesn't mean humans have vanished from the loop. Someone still has to decide what the code should do, review it for security flaws, and sign off before it ships. But the job is changing. Instead of writing every line, engineers increasingly review, correct, and direct AI output.
For a fintech platform, the shift carries weight. Exchanges live and die on uptime, security, and speed. Bugs in trading engines or wallet infrastructure can cost real money. AI-assisted code can accelerate delivery, but it also raises fresh questions about review rigor and accountability when something breaks.
The fintech pressure to ship faster
Xu framed the AI-driven engineering shift as a transformative trend in fintech — a sector where product cycles are measured in weeks, not quarters. Crypto exchanges compete on listing speed, fee structures, and new features. The faster a team can turn an idea into working code, the faster it can ship.
That pressure explains why AI adoption has moved quickly from experiment to default at some firms. If AI can draft a pull request in seconds, engineers can spend more time on architecture, risk, and the parts of the job that resist automation. Efficiency gains compound across hundreds of small changes.
But the 95% figure also cuts against a common assumption. Much of the public debate around AI and jobs focuses on customer service or content creation. Xu's disclosure points to something quieter and more technical: the automation of the code pipeline itself, inside a regulated financial business.
What happens to the engineers
The evolving role of human engineers sits at the center of this. If AI writes the first draft of nearly every pull request, the value of a developer shifts toward judgment — spotting a subtle bug, knowing which shortcut will haunt you in six months, understanding the business logic behind a feature.
That's a different skill set from raw coding speed. It rewards seniority, domain knowledge, and skepticism. It also means junior engineers may learn the craft differently, reviewing AI output rather than writing from a blank file.
OKX hasn't detailed how it retrains or reassigns staff around this workflow. Xu's statement didn't include headcount figures, timelines, or which specific AI tools the company uses. Those gaps matter for anyone trying to judge whether 95% is a milestone or a marketing line.
The accountability question
Code generation is one thing; code responsibility is another. When an AI-authored pull request introduces a flaw, who owns it? The engineer who approved it, presumably — the same as always. But if reviewers are skimming more and writing less, the safety net thins.
OKX operates in a sector where hacks and outages make headlines. The efficiency case for AI is clear. The risk case is less discussed, and the 95% figure doesn't tell us how the company is managing it.
What's next is simple to state and hard to answer: whether other exchanges and fintech firms follow OKX's lead, and whether regulators start asking how AI-written code gets audited. Xu's number puts the question on the table. The industry hasn't answered it yet.
