Block CTO Dhanji Prasanna: Building the AI-First Enterprise with Goose, their Open Source Agent
🎯 Summary
Podcast Summary: Block CTO Dhanji Prasanna on Building the AI-First Enterprise with Goose
This 59-minute episode of Training Data features Block CTO Dhanji Prasanna discussing Block’s aggressive pivot toward becoming an “AI-First” enterprise, centered around their open-source, extensible AI agent, Goose.
1. Focus Area
The discussion centers on the practical application of AI Agents in a large enterprise setting, contrasting traditional Machine Learning (ML) with modern Generative AI/Deep Learning. Key themes include organizational transformation, the role of tool use and agent middleware, and the development and philosophy behind Goose. The conversation also touches upon Block’s broader activities, including Square, Cash App, and their Bitcoin initiatives (Bitkey, RIG mining).
2. Key Technical Insights
- Goose as Agent Middleware: Goose is positioned as the “arms and legs” for the LLM “brain in a jar,” providing the orchestration layer necessary for AI to act in the real world by connecting to existing digital systems.
- Model Context Protocol (MCP): Goose is built using the MCP, a formalized standard for creating wrappers around existing tools/capabilities, allowing the agent to connect to and orchestrate diverse systems (e.g., Snowflake, Looker, Gmail).
- Learning Through Doing & Recipes: Block avoids over-engineering tool compatibility, preferring to let Goose learn workflows organically. Successful user workflows can be “baked” into shareable scripts called recipes.
3. Business/Investment Angle
- AI as Strength, Not Threat: Prasanna views AI not as a disruption that will usurp Block, but as a necessary technology to embrace, consistent with Block’s history of adopting innovations (like the original card reader or blockchain).
- Organizational Centralization: Block transitioned from a siloed General Manager (GM) structure to a centralized, functional organization to drive engineering excellence and unify AI policy across Square, Cash App, and Tidal.
- Agent Utility for Non-Engineers: A major unexpected business value is non-technical staff (sales, finance) using Goose to build their own software dashboards and reporting tools, unlocking significant productivity.
4. Notable Companies/People
- Dhanji Prasanna (Block CTO): Drove the central mandate for Block to become AI-first, advocating for centralized investment in AI transformation.
- Jack Dorsey (Block Founder): Fully aligned with Prasanna on the necessity of AI investment; credited with driving the company’s strong focus on Bitcoin.
- Brad Axen (Goose Developer): The engineer who developed the initial thesis and prototype for Goose, which was subsequently ringfenced and scaled.
5. Future Implications
The conversation suggests a future where AI agents, leveraging protocols like MCP, become the primary interface for interacting with enterprise software, moving beyond simple chat interfaces. Block is aiming for recursive self-improvement, with the goal of Goose rewriting 100% of its own codebase in future releases. The most advanced use cases hint at agents operating with high autonomy, monitoring communications and proactively developing features or managing schedules.
6. Target Audience
This episode is highly valuable for AI/ML Engineers, CTOs, Enterprise Software Architects, and Technology Strategists interested in the practical implementation, organizational structuring, and security considerations of deploying autonomous AI agents within a large, established technology company.
🏢 Companies Mentioned
💬 Key Insights
"As I told you before, in AI-first teams, it's pretty much all entirely vibe coded. So Goose itself, every PR that is open is written by Goose."
"Do you have a, and if you can't share us, okay, but I'm curious if you even look at how much of the code base is being written by Goose or being written by AI today and do a guess for how that might evolve over time. Yeah, we are measuring it. And there's different numbers for different teams. Like I said, engineers in the most engaged engineers with Goose probably generate about 30 to 40% of the code they write in existing legacy code bases..."
"Where I think humans are called for is in the higher-level architectural design, in understanding race conditions and coordinating, orchestrating across multiple systems in a topology. Those kinds of things we definitely need people in the LLM for."
"Where they do fail is understanding how to call proprietary APIs, because these are not in the training set often. And especially if you have very complex proprietary frameworks, then they can struggle to reason about them. And for that, you definitely need manual intervention."
"I think that it's a difficult proposition the more complex the code goes. And this is why I think it's more effective to vibe code these smaller tools like dashboards and reports and interactive kind of systems on a per individual basis rather than make massive changes to 10 million line code bases."
"I write code every day, and it's all through Goose. It's all through vibe code or through some of the other AI agents when I'm evaluating how good they go. So I very rarely manually write code."