BTCBTC$64,926+1.62%|
ETHETH$1,817.4+2.34%|
USDTUSDT$0.99962+0.02%|
BNBBNB$593.6600+1.81%|
USDCUSDC$0.99975+0.01%|
XRPXRP$0.96498+3.00%|
SOLSOL$71.8800+2.59%|
TRXTRX$0.30366-1.01%|
FIGR_HELOCFIGR_HELOC$1.0570-0.88%|
HYPEHYPE$66.2800+2.17%|
ZECZEC$888.1800+0.59%|
XMRXMR$493.8700+3.24%|
DOGEDOGE$0.05899+2.10%|
USDSUSDS$0.99991+0.01%|
LEOLEO$9.1600+0.72%|
WBTWBT$62.8200+1.31%|
LINKLINK$9.1800+2.12%|
ADAADA$0.15294+3.35%|
USDEUSDE$0.99967+0.02%|
XLMXLM$0.13916+2.54%|
DAIDAI$0.99994+0.01%|
BCHBCH$209.0400+3.21%|
NEARNEAR$3.1600+2.23%|
LTCLTC$44.8500+1.24%|
CCCC$0.08566+10.27%|
AVAXAVAX$7.2000+2.26%|
XAUTXAUT$3,866.61+1.26%|
UNIUNI$5.1500+4.54%|
USD1USD1$0.99865+0.07%|
GRAMGRAM$1.0590+1.40%|
BTCBTC$64,926+1.62%|
ETHETH$1,817.4+2.34%|
USDTUSDT$0.99962+0.02%|
BNBBNB$593.6600+1.81%|
USDCUSDC$0.99975+0.01%|
XRPXRP$0.96498+3.00%|
SOLSOL$71.8800+2.59%|
TRXTRX$0.30366-1.01%|
FIGR_HELOCFIGR_HELOC$1.0570-0.88%|
HYPEHYPE$66.2800+2.17%|
ZECZEC$888.1800+0.59%|
XMRXMR$493.8700+3.24%|
DOGEDOGE$0.05899+2.10%|
USDSUSDS$0.99991+0.01%|
LEOLEO$9.1600+0.72%|
WBTWBT$62.8200+1.31%|
LINKLINK$9.1800+2.12%|
ADAADA$0.15294+3.35%|
USDEUSDE$0.99967+0.02%|
XLMXLM$0.13916+2.54%|
DAIDAI$0.99994+0.01%|
BCHBCH$209.0400+3.21%|
NEARNEAR$3.1600+2.23%|
LTCLTC$44.8500+1.24%|
CCCC$0.08566+10.27%|
AVAXAVAX$7.2000+2.26%|
XAUTXAUT$3,866.61+1.26%|
UNIUNI$5.1500+4.54%|
USD1USD1$0.99865+0.07%|
GRAMGRAM$1.0590+1.40%|
News/AI & Blockchain
AI & Blockchain

From 40% to Nearly All of It: Inside Coinbase’s Sprint to AI-Written Code

BitnxtWritten by : Bitnxt
July 15, 20265 min read
From 40% to Nearly All of It: Inside Coinbase’s Sprint to AI-Written Code
Coinbase says AI now writes or assists with up to 100% of its code, marking one of the fastest AI transformations in finance while raising questions about security, oversight, and workforce changes.

A little over a year ago, artificial intelligence wrote a negligible share of the software running America's largest crypto exchange. Today, Coinbase says AI has a hand in almost all of it. According to the company's head of platform, Rob Witoff, somewhere between 95% and 100% of Coinbase's code is now written “by or with” large language models — and effectively every employee uses AI daily. It's one of the most aggressive corporate AI adoptions yet disclosed, and it doubles as a real-time experiment in what happens when a regulated financial company hands most of its coding to machines.

A staggering acceleration

The speed of the shift is the real headline. In the first quarter of 2025, AI-generated code made up just 5.7% of Coinbase's codebase. By late 2025 it had crossed 50%; by early 2026, roughly 80%; and in February the company still pegged the figure around 40% by one internal measure. Now it's brushing up against 100%. CEO Brian Armstrong drove the change relentlessly, at one point publicly setting a target to get AI past 50% of daily code and — by his own later admission, “heavy-handedly” — even firing engineers who refused to adopt the tools.

The productivity claims are equally eye-catching. Witoff says most Coinbase engineers now run five to ten AI agents simultaneously, and that these agents collectively perform the coding work of about 1,200 employees. By 2030, he speculates, Coinbase's AI agents could do the equivalent labor of 100,000 people. Notably, the company says its AI spending has stayed roughly flat even as usage exploded.

The human cost behind the numbers

This transformation didn't happen in isolation — it's inseparable from job cuts. In May 2026, Coinbase laid off about 700 people, roughly 14% of its workforce. Armstrong framed it as a return to “the speed and focus of our startup founding, with AI at our core.” Witoff acknowledged that junior development roles were hit hard, though the cuts spanned marketing, legal, customer support, and compliance too. The company has reorganized around smaller, more senior teams, where two or three people now do work that once required ten or more.

Coinbase is far from alone. Block's Jack Dorsey cut 40% of his workforce, Crypto.com trimmed 12%, and Kraken, Gemini, Messari, and Dune all cited AI-related efficiencies in their own layoffs. A pattern is emerging across the industry: AI as both the justification and the mechanism for leaner headcounts — which is precisely what makes some observers uneasy.

You might also like: A Quiet Door Opens: Coinbase Eases China Sign-Ups, and COIN Stock Takes Notice

Not all code is created equal

The most important nuance in Coinbase's disclosure is that the “95-100%” figure hides a wide spectrum of how AI is actually used. The company draws a sharp line based on risk. For low-stakes work like internal prototyping, the workflow is now essentially 100% automated. But for the crown jewels — core cryptography and anything touching private-key management — humans remain firmly in charge. As Witoff put it, industry-leading cryptographers still review that code meticulously, one line at a time.

In other words, letting an AI generate a new dashboard feature carries limited downside; letting it loose on the code that secures billions in customer funds does not. Coinbase says it maps this using a “repository sensitivity matrix” built with its security and privacy teams, and that mandatory human review of AI-produced code remains a firm requirement. The engineer's role, increasingly, is shifting from writing code to supervising the agents that write it — from producer to reviewer.

The risks nobody should wave away

Moving from 40% to near-100% AI-assisted code in a matter of months is blisteringly fast even by tech standards, and it raises legitimate questions. In a regulated financial context, a software bug isn't just a bad user experience — it can be a security breach or a compliance event with real consequences. Critics also warn of a subtler danger: if engineers increasingly review rather than write, their deep understanding of the systems may erode over time, making it harder to catch the vulnerabilities that AI can silently introduce. Coinbase itself notes AI raises bug risks in some areas even as it accelerates others.

There's a data-ethics dimension too. As one industry figure pointed out, much of the AI that now writes Coinbase's code was trained on human-created work — and little of the value generated flows back to those original creators, reviving the broader debate about data ownership in the AI era.

Why it matters beyond Coinbase

Coinbase's claim is more than a productivity milestone; it's a signal that major financial-infrastructure companies are redesigning software development itself around AI. For investors, the appeal is obvious: a firm that can build and maintain products with far fewer people enjoys a fundamentally different cost structure and higher operating leverage. If Coinbase can sustain its pace with leaner teams, competitors from Robinhood to traditional exchanges will face pressure to match it. The template — smaller senior teams supervising fleets of AI agents — could ripple across the entire sector.

The bottom line

Coinbase has effectively become a live stress test for AI-native software development at scale, in one of the least forgiving environments imaginable: regulated finance, where trust and security are the product. The efficiency gains are real and the strategic logic is clear. But the honest verdict is that the experiment isn't finished. The decisive question isn't whether AI can write 95% of the code — Coinbase has shown it can — but whether that speed can be maintained over years without compromising the security, reliability, and trust that a crypto exchange ultimately runs on. The answer will take longer than a quarter to reveal itself.

 

Sources: Comments from Coinbase’s Rob Witoff and CEO Brian Armstrong via Cointelegraph; reporting from Crypto Briefing, FinanceFeeds, The Block, and Decrypt; industry context from OpsLevel and Y Combinator (Sept 2025–July 2026). This article is for informational purposes only and is not investment advice.

#Coinbase#AI#Artificial Intelligence#AI Coding#Large Language Models
Bitnxt

Author

Bitnxt

Crypto News Writer · Bitnxt

Covering the latest developments in cryptocurrency, blockchain technology, and digital asset markets.

More AI & Blockchain News

Share:

Bitnxt uses cookies to run the site, understand how it is used, and measure ads. You can accept or decline below, and change your choice any time in our Privacy Policy.