Coinbase's AI Push Forces Engineers to Evolve or Exit

Coinbase’s CEO, Brian Armstrong, has taken a firm stance in pushing for increased adoption of artificial intelligence within the company’s software development process. As of now, approximately 40% of the daily code written at the cryptocurrency exchange is generated by AI, according to Armstrong, who is determined to raise that figure to over 50% by October 2025 [1]. In a move that has drawn attention across the tech and financial sectors, Armstrong mandated that all employees begin using AI coding tools like GitHub Copilot and Cursor last month. Those who did not comply with the directive faced immediate consequences, including termination [2].

The AI mandate, which Armstrong described as a strategic imperative, was communicated through an internal Slack channel and followed by an aggressive onboarding process. Employees were expected to integrate the tools within a week. Armstrong explained that non-compliant engineers who could not justify their inaction were asked to provide explanations, and some were ultimately let go [1]. He emphasized the importance of AI in transforming engineers into “AI-natives,” a term used to describe professionals who are proficient in leveraging AI as part of their daily workflows [3].

Coinbase’s approach to AI integration is underpinned by a framework that includes metrics from the DevOps Research and Assessment (DORA) model, as well as AI-specific signals such as token usage and the proportion of AI-generated code. The company is closely tracking lead-time-to-change, deployment frequency, and the share of AI-generated code to measure progress. According to internal data, the percentage of AI-generated code at Coinbase has more than doubled since April 2025 [2]. Armstrong noted that while AI is being used across multiple areas—such as frontend development, testing, and new project creation—not all code can or should be AI-generated due to security and sensitivity concerns [1].

The CEO also highlighted that AI is not intended to replace a significant number of the company’s 4,200 employees. Instead, it is meant to augment productivity and streamline operations. Armstrong pointed to success stories in which engineers have been able to refactor, upgrade, or build new codebases in days instead of months thanks to AI tools. However, the use of AI also introduces risks, such as the potential for bugs, which Coinbase is addressing through a repository sensitivity matrix developed in collaboration with its security and privacy teams [1].

Coinbase’s AI-driven approach aligns with broader industry trends. For instance, Google and Microsoft have reported significant AI code generation in their own development processes, with Microsoft projecting that the share of AI-generated code could reach 50% within the next two years [3]. Academic research further supports these trends, showing that AI-generated code already accounts for 30.1% of functions in the United States, with similar figures in Germany and France [3]. As companies like Coinbase continue to embed AI into their development pipelines, the balance between human and machine contributions is shifting, reshaping not only software development but also the broader technology landscape [3].

Source:

[1] Brian Armstrong says about 40% of Coinbase's daily code... (https://www.theblock.co/post/369460/brian-armstrong-coinbase-ai)

[2] Coinbase's Brian Armstrong Says AI Writes 40% Of Its Code (https://cointelegraph.com/news/coinbase-says-ai-writes-nearly-half-of-its-code)

[3] Coinbase Targets Majority AI Code by October (https://beincrypto.com/coinbase-ai-code-push-amid-global-adoption/)