Coinbase has quietly crossed a threshold that would have seemed implausible just five months ago. According to a report from WuBlockchain citing an interview with Rob Witoff, Coinbase’s Head of Platform, 95% to 100% of the company’s code is now written by or with the assistance of AI and large language models.
That is a sharp jump from the 40% figure Witoff estimated back in February. The speed of the shift is even more striking when paired with another detail: effectively 100% of Coinbase employees use AI daily. Most engineers are running between five and ten AI agents at any given time. Together, those agents perform the coding work equivalent to roughly 1,200 full‑time employees, according to Witoff.
From Efficiency Play to Structural Shift
Coinbase’s internal AI usage is no longer an experiment. The company has moved from tinkering with code completion tools to a model where large language models generate the overwhelming majority of its software. The projection that AI agents could be doing the work of 100,000 employees by 2030 is the kind of figure that rewrites headcount strategies across the industry.
For a publicly traded crypto exchange that also runs the Base Layer‑2 network, the implications cut two ways. On one side, slashing the marginal cost of producing code could dramatically improve margins and accelerate product development. On the other, the defense‑grade quality assurance required for custody, trading engines, and on‑chain settlement suddenly depends on code that no single human wrote or fully understands.
While Coinbase is automating its own development, the broader crypto industry is exploring ways to embed AI into decentralized infrastructure. Recent partnerships, such as UXLINK’s integration with Origins Network , aim to power scalable AI‑driven applications using distributed computing resources. That parallel effort highlights a growing divergence: some firms are building AI for their own efficiency, while others are building decentralized infrastructure for AI workloads.
What This Means for Crypto Software Practices
The move also puts pressure on other exchanges and infrastructure providers. If Coinbase can deploy AI agents that simulate the output of a thousand‑plus developers, competitors who rely mostly on traditional coding may see their relative pace of innovation shrink. Tight developer market conditions in crypto make the proposition even more attractive. Junior coding roles, already under scrutiny in tech, face an existential question inside financially disciplined crypto firms.
Yet the transition is not frictionless. AI‑generated code tends to introduce subtle bugs, especially around edge cases that a senior engineer might catch during review. For a platform handling billions in customer assets and real‑time settlement, even a single deployment error carries non‑trivial risk. The security model that Coinbase has built over the last decade now has to adapt to a codebase whose lineage is increasingly synthetic.
Developers, Data, and the Long‑Term View
Coinbase’s embrace of AI‑assisted coding also intersects with the crypto industry’s growing appetite for verifiable computation and decentralized data storage. As detailed in a recent Filecoin price analysis , networks offering AI‑related storage are starting to be re‑assessed by traders who see enterprise AI adoption as a long‑term demand driver. The deepening relationship between AI and software creation could eventually shift demand from code frameworks toward data curation and model reliability.
It lands at a moment when developer activity across top blockchains remains heavily skewed toward human contributions. While Coinbase is an outlier in its scale of internal AI usage, the wider industry is still largely shaped by open‑source communities and manual workflows. How quickly that status quo changes may determine which platforms gain the strongest engineering edge over the next cycle.
For now, the numbers from Coinbase serve less as a victory lap and more as a signal to other crypto firms. The gap between those who can integrate AI deep into their software supply chain and those who cannot is widening fast. What remains untested is whether that speed advantage holds under the scrutiny of high‑stakes financial infrastructure, or introduces fragility that only becomes visible at scale.
