AI-Assisted Code as the New Normal
AI-assisted code development is the practice of embedding large language models into the enterprise development workflow so that most new and modified code is either generated outright or written with real-time machine suggestions, turning AI from a side tool into a default part of how software is produced at scale. Coinbase’s recent numbers show this shift is no longer hypothetical. Platform head Rob Witoff says that 95% to 100% of the company’s code is now written or assisted by large language models, compared with 40% in February, a leap that highlights how quickly AI code generation can become standard once it is integrated deeply into daily engineering work.
From 40% to Almost Everything: A Deliberate Sprint
The striking part of Coinbase’s trajectory is not that it uses AI for code, but how fast it went from minority to near-total adoption. Moving from 40% of code written with AI assistance in February to between 95% and 100% by mid-July is less an incremental improvement than a wholesale redesign of the development process. This did not happen in a vacuum. In May, the company cut 700 roles, with CEO Brian Armstrong arguing that AI had already changed the pace of work and that the company needed to "return to startup speed" with AI at the core of that rhythm. That linking of workforce changes to AI usage sends a clear message: AI is not an optional developer productivity tool; it is part of the operating model.
What Coinbase’s Shift Reveals About Enterprise Workflows
Coinbase’s rapid transformation shows how an enterprise development workflow can be rebuilt around large language models in months rather than years. When leadership frames AI as central to "work pace" and demands startup-level speed, engineers are pushed to make LLM code assistance the default path for writing features, tests and maintenance tasks. The result is that manual-first coding becomes the exception. This is exactly how AI code generation moves from novelty to infrastructure: it is embedded in tooling, backed by executive pressure, and aligned with hiring and restructuring decisions. The message to other large organizations is blunt. If you still treat LLMs as sidecar experiments, you will move at a different speed than peers who rebuild their pipelines around AI suggestions and automation.
Impact Beyond Engineers: Users and Compliance Pressures
It is tempting to view AI-assisted coding as purely an internal story, but Coinbase’s timing suggests wider forces. As regulatory transition periods like MiCA’s July 1 cutoff end, officials warn that user migration could increase compliance pressure on virtual asset service providers and make it harder for licensed platforms to attract and onboard new customers efficiently. Faster, AI-accelerated development is one way to respond: teams can ship features to support new rules, reporting needs and risk controls more quickly than they could with a traditional workflow. For ordinary users, this should mean more frequent product updates and smoother adaptation to changing rules. The hidden bet is that AI-enabled speed will help compliant platforms keep up with both regulation and user demand.
The Enterprise Lesson: Make AI the Default, Not the Add-On
Coinbase’s journey from 40% to near-universal AI-assisted code in a few months is not a quirky outlier; it is a roadmap for large engineering organizations that want to move faster without waiting for perfect AI maturity. The company tied AI adoption directly to its strategy, workforce decisions and expectations about delivery speed, showing that meaningful gains come when AI code generation is woven into every layer of the enterprise development workflow. The conclusion is clear: if AI remains a separate tool that only some developers use, you will not see Coinbase-level shifts. To capture the same momentum, enterprises need to commit to LLM assistance as the default way code gets written, reviewed and updated—and accept that the bigger risk now is moving too slowly, not experimenting too much.
