How Block is becoming the most AI-native enterprise in the world | Dhanji R. Prasanna
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this captivating episode, Dhanji R. Prasanna, CTO of Block, shares his journey from writing an "AI manifesto" to Jack Dorsey that led to his promotion, to transforming Block into one of the most AI-native large companies in the world. (05:26) Dhanji reveals how their internal open-source agent called Goose is saving employees 8-10 hours weekly, discusses the organizational changes that made this transformation possible, and provides insights into the future of AI-augmented work.
- Main themes: AI transformation at enterprise scale, organizational structure's impact on technology adoption, the development and impact of Block's open-source AI agent Goose, and practical lessons for companies looking to become more AI-native
Speakers
Dhanji R. Prasanna
Chief Technology Officer at Block (formerly Square), where he manages more than 4,000 engineers. Under his leadership, Block has become one of the most AI-native large companies in the world. Before becoming CTO, Dhanji wrote an "AI manifesto" to CEO Jack Dorsey that sparked a company-wide transformation and led to his promotion to CTO.
Lenny Rachitsky
Host of Lenny's Podcast and author of Lenny's Newsletter, focusing on product management, growth, and startup advice for ambitious professionals.
Key Takeaways
Organizational Structure Drives Technology Adoption
Dhanji emphasizes that Conway's Law - "you ship your org structure" - is incredibly powerful when implementing AI transformation. (09:09) Block moved from a GM structure where different business units (Square, Cash App, Afterpay) operated independently with separate engineering teams, to a functional structure where all engineers report to one engineering leader. This change was essential for driving AI adoption company-wide because it created singular focus and allowed for shared tools, policies, and technical strategy across all teams.
Start with Leadership Using the Tools Daily
The most effective way to drive AI adoption throughout an organization is for leadership to use the tools themselves every single day. (54:04) Dhanji notes that Jack Dorsey, himself, and the entire executive team use Goose regularly, which gives them deep understanding of the tools' strengths, weaknesses, and ergonomics. This hands-on approach is far more effective than reading articles and then trying to get teams to follow suit - you need to "feel it, use the product yourself" to understand how to apply it organizationally.
Non-Technical Teams Benefit Most from AI Tools
Surprisingly, the teams seeing the most dramatic productivity gains aren't the engineering teams, but non-technical teams using AI agents and programming tools to build their own solutions. (49:55) Enterprise risk management teams are building entire self-service systems, compressing weeks of work into hours, rather than waiting for internal app teams to put requests on quarterly roadmaps. This represents a fundamental shift in how work gets done across organizations.
Focus on Human Taste and Judgment Over Technical Perfection
Code quality has almost nothing to do with product success, as demonstrated by YouTube's early architecture storing videos as blobs in MySQL while being far more successful than Google Video, which was technically superior. (62:03) The key insight is that human taste and judgment matter more than technical perfection - AI needs to be anchored by human taste to avoid "AI slop" and ensure outputs are meaningful, tasteful, and valuable to real people solving real problems.
AI Should Work Continuously, Not Just During Human Hours
The future of AI-augmented work involves agents working autonomously for hours rather than minutes, including overnight and weekends when humans aren't available. (32:42) Rather than the current "ping pong" model where you ask for something and wait 3-4 minutes for a half-baked response, Dhanji envisions describing multiple experiments in detail, going to sleep, and waking up to find all experiments built and ready for evaluation - allowing teams to experiment with multiple approaches simultaneously rather than choosing just one path.
Statistics & Facts
- Engineering teams using Goose daily are reporting 8-10 hours saved per week, with Block trending toward 20-25% of manual hours saved across the entire company (not just engineering teams). (16:37) This metric is self-reported but validated through multiple check metrics including PRs, feature throughput, and data scientist analysis.
- Block has over 4,000 engineers under Dhanji's leadership, making it one of the largest engineering organizations to undergo such a comprehensive AI transformation. (01:37)
- Goose's median session length is 5 minutes with an average of 7 minutes, but the team is working to push this to hours of autonomous work rather than short interactive sessions. (33:33)
Compelling Stories
The AI-Watching Engineer: One of Block's engineers has taken AI adoption to an extreme level by having Goose continuously watch his screen and listen to his conversations. (28:52) When he discusses potential features with colleagues over Slack or email, he'll find hours later that Goose has already attempted to build that feature and opened a pull request on Git, demonstrating the potential for truly proactive AI assistance.
The Receipts Solution: Dhanji shared a personal example of using Goose to solve a real problem - organizing therapy receipts for his son with additional needs. (55:55) Goose figured out how to convert various receipt formats (PDFs, screenshots) into HTML, organize them in Apple Notes for seamless phone syncing, and enable easy sharing with his wife for insurance claims - a creative solution using AppleScript that Dhanji never would have thought of himself.
Thought-Provoking Quotes
Dhanji R. Prasanna
"A lot of engineers think that code quality is important to building a successful product. The two have nothing to do with each other." (62:03)
Context: Discussing counterintuitive lessons about building products, using YouTube's early architecture as an example of how technical perfection doesn't correlate with product success.
Dhanji R. Prasanna
"All these LLMs are sitting idle overnight and on weekends while humans aren't there. Like, there's no need for that. They should be working all the time." (33:42)
Context: Describing his vision for the future of AI-augmented work, where agents work autonomously for extended periods rather than just short interactive sessions.
Dhanji R. Prasanna
"The truth is the value is changing every day, so you need to ride that wave along with it." (17:59)
Context: Addressing skepticism about AI productivity gains, emphasizing that companies need to stay adaptable as AI capabilities rapidly evolve.
Dhanji R. Prasanna
"If you're not waking up in the morning feeling energized about what you're going to do that day in your professional life, then change something." (82:04)
Context: Sharing his favorite life motto during the lightning round, advocating for taking control of one's career satisfaction.
Strategies & Frameworks
The Model Context Protocol (MCP) Integration Strategy
Block's approach to AI agent development centers around the Model Context Protocol, which creates formalized wrappers around existing enterprise tools like Salesforce, Snowflake, and SQL databases. (22:38) This allows AI agents to orchestrate across multiple systems seamlessly, turning chatbots into actionable agents with "arms and legs" in the digital world.
Practical Example: A marketing professional can ask Goose to create a complete marketing report, and it will automatically write SQL queries to pull data from Snowflake, perform analysis in CSV format using Python, generate interactive charts with JavaScript libraries, compile everything into a PDF or Google Doc, and even email the final report.
Identify Core Enterprise Systems
Catalog all the tools your organization uses daily - from databases and CRM systems to communication platforms and file storage. Each of these can potentially become an MCP endpoint that AI agents can manipulate.
Create MCP Wrappers
Build simple API wrappers for each system that expose key functionalities to AI agents. The beauty is that with just a few lines of code, entire systems become orchestratable by AI overnight.
Enable Cross-System Workflows
Train your team to think in terms of multi-system workflows rather than single-tool tasks. Instead of manually moving data between systems, describe the end goal and let the AI agent handle the orchestration.
Functional Organization for AI Transformation
Moving from a GM structure (where business units operate independently) to a functional structure (where all engineers report to one leader, all designers to another) is crucial for successful AI adoption at scale. (09:09) This creates the singular focus needed to drive deep technology initiatives like AI transformation company-wide.
Practical Example: Instead of having separate engineering teams for Square, Cash App, and other business units each making independent AI tool decisions, Block now has unified engineering standards, shared AI tools like Goose, and common policies that enable knowledge sharing and coordinated AI adoption.
Assess Current Organizational Silos
Identify where different business units or teams are making independent technology decisions that prevent company-wide AI adoption. Look for duplicated effort and inconsistent tool choices.
Consolidate Technical Leadership
Create singular reporting structures for key functions like engineering and design, ensuring one leader can drive consistent AI strategy across all business units rather than having to negotiate with multiple independent teams.
Establish Shared Standards and Tools
Implement common AI tools, development practices, and policies across the entire organization. This enables best practices to spread quickly and prevents teams from reinventing the wheel in isolation.
Similar Strategies
Agile Methodology
Agile is similar to the organizational changes Danji Prasana discusses in terms of its focus on flexibility and continuous improvement. It emphasizes collaboration, iterative development, and quick adaptation to change, similar to how Block moved from a GM to a functional structure to enhance tech focus and efficiency.
Lean Startup Approach
The Lean Startup methodology is akin to Block's approach of starting small and iterating quickly. It involves rapid experimentation, customer feedback, and continuous improvement, mirroring the hack week ideas that grew into major products like Cash App.
Additional Context
- Conway's Law: The principle that organizations design systems that mirror their communication structures, which proved crucial in Block's AI transformation success.
- Model Context Protocol (MCP): An open protocol developed by Anthropic that allows AI agents to interact with external tools and systems, forming the foundation of Goose's capabilities.
- Goose: Block's open-source AI agent that can perform tasks ranging from organizing photos to writing software, built on the MCP protocol and available for free download.
- Recurring theme: The importance of starting small with AI experiments before scaling, as demonstrated by both Goose and Cash App beginning as hack week projects.
Key Takeaways Table
| Key Takeaway | Breakdown |
|---|---|
| Organizational Structure Drives AI Success | Moving from GM to functional structure enables unified AI strategy and tool adoption across business units |
| Leadership Must Use AI Tools Daily | Executives using tools themselves is more effective than reading about AI and directing others to adopt it |
| Non-Technical Teams Benefit Most | Risk management, legal, and support teams see dramatic gains building their own AI-powered solutions |
| Human Taste Anchors AI Output | Technical perfection matters less than human judgment in creating meaningful, valuable solutions |
| AI Should Work Continuously | Future involves agents working autonomously for hours/overnight rather than short interactive sessions |
Critical Analysis
While Dhanji's insights about AI transformation are compelling, several factors may limit their universal applicability. Block's success with AI tools like Goose benefits from their status as a technology company with significant engineering resources and an existing culture of experimentation. The 8-10 hours of weekly savings metric, while impressive, is self-reported and may not account for the time invested in learning, maintaining, and troubleshooting these AI tools. Additionally, the recommendation to reorganize from GM to functional structure represents a massive organizational undertaking that could be disruptive for many companies, potentially outweighing short-term AI productivity gains. The emphasis on non-technical teams building their own software solutions, while empowering, could also create governance, security, and maintenance challenges that aren't fully addressed in the discussion.
Books & Articles Mentioned
- There are no books mentioned directly related to the technology or AI strategies discussed in the episode. However, the guest, Danji Prasana, recommends reading fiction and classics such as The Master and Margarita by Mikhail Bulgakov, and Tennyson's poetry.
Products, Tools & Software Mentioned
- Goose – A general-purpose open-source AI agent developed by Block, used for automating tasks and improving productivity.
- Figma Make – A tool for creating prototypes and apps using AI-based vibe coding.