Block CTO Dhanji Prasanna: Building the AI-First Enterprise with Goose, their Open Source Agent
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
Block CTO Dhanjal Prasanna shares insights from Block's comprehensive AI transformation journey, centered around their open-source AI agent Goose. The conversation explores how Block reimagined itself as a technology company leveraging AI across all functions, from engineering productivity to customer-facing products. (01:24) Prasanna reveals that Goose is now writing the vast majority of new code for its own codebase, demonstrating the recursive potential of AI agents. The discussion covers practical implementation strategies, from organizational restructuring to enable AI adoption to measuring real impact through manual hours saved - currently targeting 25% by year-end. (32:08)
- Main Theme: How enterprises can successfully transform into AI-native organizations through strategic agent deployment, organizational design changes, and focus on unlocking utility rather than chasing capabilities.
Speakers
Dhanjal Prasanna
CTO at Block with over 10 years of involvement with the company, including his first GitHub commit dating back to 2011. Prasanna led Cash App's engineering team from 10 to over 200 engineers and was instrumental in Block's organizational transformation from GM structure to functional organization. He spearheaded Block's AI transformation initiative after writing a comprehensive AI manifesto to founder Jack Dorsey, advocating for centralized AI investment across the entire company.
Key Takeaways
Embrace Autonomous Agent Workflows Over Traditional Tool Integration
Rather than trying to engineer tools to be "AI-friendly," Block discovered that letting AI agents learn organically produces superior results. (28:19) Goose figures out how to use existing systems in surprising ways that humans wouldn't consider, often faster than manual approaches. This philosophy of "not over-engineering" and allowing agents to develop their own workflows through experimentation has proven more effective than attempting to create perfect integrations upfront. The key insight is that agents can adapt to existing infrastructure rather than requiring infrastructure to adapt to agents.
Implement Organizational Design That Enables AI Adoption
Block's transformation from a GM structure to a functional organization was critical for AI success. (10:37) Prasanna explains that centralization allowed them to drive engineering excellence, unify policies, and create the depth needed for AI transformation. In an era of rapid technological shifts, having singular organizational focus becomes crucial for staying competitive. This structural change enabled Block to treat all capabilities - from taking payments to creating GitHub issues - as unified functions that AI agents could orchestrate across.
Measure AI Impact Through Manual Hours Saved
Block tracks AI effectiveness using a specific metric: manual hours saved by Goose on a weekly basis. (32:08) This metric started at 0% and is targeting 25% by year-end, with engineers reporting 8-10 hours saved per week. The metric combines qualitative and quantitative signals to provide a comprehensive view of AI impact. This measurement approach focuses on practical value delivery rather than abstract productivity metrics, giving organizations a concrete way to assess their AI investments.
Focus on Utility Over Capability When Deploying AI
Prasanna emphasizes that many companies rush toward AI without understanding how to unlock utility from it, leading to incorrect conclusions about AI's value. (56:56) The key is identifying what LLMs are genuinely good at - general tasks and workflow automation - rather than expecting them to excel at highly specialized knowledge work. Success comes from applying AI to eliminate "work about work" and manual drudgery, while recognizing that deep domain expertise still requires human involvement.
Prepare for Swarm Intelligence as the Next Evolution
The future of AI coding capability lies not in single agents but in swarm intelligence - leveraging multiple instances working together. (39:38) Prasanna believes the competition won't be about whether open source models match closed models, but whether you can run hundreds of smaller models that collectively outperform any single large language model. This approach could enable building complex applications like Cash App through coordinated agent collaboration, fundamentally changing how we think about software development scale and capability.
Statistics & Facts
- Block's AI transformation is targeting 25% manual hours saved by the end of the year, with engineers currently reporting 8-10 hours saved per week using Goose and other AI tools. (32:08)
- Block's open source Android technologies run on close to 4 billion mobile devices worldwide, demonstrating the company's massive scale in open source contribution. (36:34)
- In AI-first teams at Block, engineers generate 30-40% of their code through AI in existing legacy codebases, while some teams like the Goose team have every PR written entirely by AI. (50:33)
