Coinbase held a time-boxed event where 100+ engineers used an AI tool to simultaneously submit PRs for trivial fixes. This created a transformational moment, breaking inertia, proving the tool's value, and generating massive, visible momentum for adoption across the entire organization.
Coinbase invented a role called the "Super Builder" whose sole job is to create more super builders. This person focuses exclusively on building internal AI tools and workflows that accelerate the entire engineering organization, acting as a powerful force multiplier for developer productivity.
To overcome skepticism in a large engineering organization, a leader must have deep conviction and actively use AI tools themselves. They must demonstrate practical value by solving real problems and automating tedious work, rather than just mandating usage from on high.
By building internal AI agents directly into Slack, their usage becomes public and visible. This visibility is key for driving adoption; seeing a bot turn a message into a PR creates a "holy shit" moment that sparks curiosity and makes others want to use the tool, creating a natural viral effect.
A Coinbase engineering director reports that after scaling AI adoption, his calendar is "almost empty." The massive reduction in coordination overhead—fewer prioritization meetings, status updates, and roadmap discussions—is a primary benefit, allowing leaders to spend more time writing code themselves.
To compress feedback cycles, Coinbase built a tool that captures live audio feedback, uses an LLM to create a structured bug report in Linear, and then triggers an internal Slack bot to immediately begin authoring a pull request. This reduces the feedback-to-fix cycle from weeks to minutes.
Vanity metrics like "AI lines of code" are misleading. Coinbase measures AI success by its impact on the end-to-end development cycle: the total time from a ticket's creation to the change landing with a user. This metric holistically captures gains and focuses the team on true velocity.
In a meta-move, Coinbase's engineering director downloaded user analytics from their AI coding tool, Cursor, and then used Cursor itself to perform a cohort analysis. This quickly identified user segments (e.g., "agent-heavy") and generated a playbook to help light users become power users.
To get skeptical engineers to adopt AI, don't focus on complex coding tasks. Instead, provide tools that automate the tedious, soul-crushing "paper cut" tasks like writing unit tests, linting, and fixing design debt. This frames AI as a tool that frees them up for more enjoyable, high-impact work.