AI-Assisted Development at Block

Angie Jones, who leads AI enablement for engineering at Block, on how you get an organisation from curiosity to habit. The strategy is tool freedom rather than a mandate: engineers pick from Claude Code, Goose, Cursor, Copilot and others, and 95% of them now use AI regularly. Progress is tracked as stages, with the largest group running a single agent outside the IDE and the next largest running three to five agents in parallel.

The lever was an AI Champions programme: 50 developers drawn from Block, Square, Cash App, Afterpay, Tidal and platform teams, each given 30% of their time for enablement work. Within three months AI-authored code rose 69%, reported time savings rose 37%, and automated pull requests went up 21 times.

The other half is preparing the codebase rather than the person. “Repo readiness” means committing context files (AGENTS.md for build and test commands and architecture, a human-facing HOWTOAI.md with setup and example workflows, plus tool-specific ones like .goosehints and CLAUDE.md), then levelling up to skills and automated review. Repo Quest gamifies this as a four-tier ladder: Locked, Novice, Adept, Artisan. For complex work they use RPI, a three-phase research, plan, implement pattern where the agent documents findings and writes an execution plan before touching code, which they credit with preventing AI drift. The stated goal is that by year end most of engineering is comfortable using orchestrators to run agents in parallel, which is the same direction Block’s own agent work has taken with Buzz and its tiered team model.