Block CTO Dhanji Prasanna: Building the AI-First Enterprise with Goose, their Open Source Agent
Block CTO Dhanji Prasanna explains how Goose, their open-source AI agent, is saving engineers 8-10 hours/week and why swarms of small models will outcompete monolithic LLMs.
- Block’s AI productivity metric targets 25% of manual hours saved by end of 2025; engineers currently report 8-10 hours/week saved.
- Goose writes the vast majority of its own codebase; the goal is 100% self-written code per release.
- Non-engineers at Block now build dashboards, reports, and functioning sites using Goose without writing code.
- Prasanna’s bet: 50-1000 instances of small, cheap open-source models running in swarms will outperform single large closed models for complex tasks.
- Block reorganized from GM business-unit silos to a centralized functional org specifically to accelerate AI adoption across Square, Cash App, and Tidal.
- Headless Goose runs in Block’s CI pipeline and auto-fixes infosec vulnerability tickets, subject to human review before any production merge.
- Square AI (public beta) lets merchants query financials conversationally; one wine bar used it to discover closing early would cut staff tips, reversing the decision.
- Block treats all internal and product capabilities as API primitives, with Goose as an agent middleware layer sitting on top — the architecture they see unlocking the next wave of utility.
2025-09-30 · Watch on YouTube
