Block CTO on Goose: AI That Opens PRs Before You Ask

How Block Became AI-Native Across 3,500 Employees

This is Dani Prasana - Block CTO overseeing 3,500 people - explaining how one of the world's largest fintech companies became AI-native. The stats are real, measured, and improving.

"8-10 hours saved per week" is self-reported, then validated. Block tracks PRs, feature throughput, and a "complicated formula" their data scientists built to distill meaning. Across the whole company (support, legal, risk - not just engineering), they're trending toward 20-25% manual hours saved. Engineering varies by codebase complexity.

One engineer has Goose watch his entire screen. He'll be discussing a feature on Slack, and hours later Goose has already tried to build it and opened a PR. It nudges him out of meetings that run over. It does things he didn't prompt for but that it thinks will help. "You have to have the stomach to be that tied into your tools."

The biggest surprise: non-technical teams. Enterprise risk management built their own self-service system. Finance, legal - they're building software tools that compressed weeks into hours. No waiting for Q2 roadmaps from internal apps teams.

Goose is entirely open source. Any model, MCP integrations, write your own extensions. Block's competitors use it. They built it on MCP (Model Context Protocol) from Anthropic and were early contributors. "Goose can write its own MCP" - it's self-bootstrapping.

Conway's Law forced the transformation. Jack Dorsey returned, 40 executives in weekly meetings, but nobody was talking about AI. Dani wrote a letter. Then they restructured from GM silos (Square, Cash App, Afterpay, Tidal run independently) to functional org (all engineers under one team). "You ship your org structure."

10 Insights From Block's AI Transformation

• 8-10 hours saved per week - Self-reported, validated by PR/throughput metrics
• 20-25% manual hours saved company-wide - Support, legal, risk, not just engineering
• Goose watches screen, opens PRs - One engineer has full screen monitoring
• Non-technical teams show most impact - Enterprise risk built their own system
• Entirely open source - Competitors use it; any model, MCP integrations
• Gling for mobile - Android accessibility API for UI test automation
• GM → Functional reorg - All engineers under one team enables AI transformation
• Conway's Law - "You ship your org structure"
• AI letter to Jack Dorsey - Dani's "manifesto" started the shift
• Green field > legacy - New codebases see aggressive gains; complex codebases less

What This Means for Enterprise AI Adoption

Block's CTO has an AI watching his screen all day, opening PRs for features discussed in Slack before he asks. This isn't sci-fi - it's production. The biggest surprise: non-technical teams (legal, finance, risk) are building their own tools, compressing weeks into hours. AI is democratizing software development inside the enterprise.