How Block is becoming the most AI-native enterprise in the world

Here's a comprehensive summary of the YouTube video transcript:

Key Concepts

AI Productivity Gains: The potential for AI to significantly increase efficiency and save time for individuals and teams.
Goose: Block's open-source, general-purpose AI agent designed to automate tasks and interact with digital tools.
Model Context Protocol (MCP): A formalized set of wrappers around existing tools and capabilities that allows LLMs to manipulate them.
Functional vs. GM Structure: Organizational structures where functional teams (e.g., all engineers report to one head of engineering) are contrasted with General Manager (GM) structures where business units operate more independently.
Conway's Law: The principle that an organization's communication structure dictates the design of the systems it builds.
Code Quality vs. Product Success: The argument that high code quality is not a prerequisite for a successful product.
Controlled Chaos: A management approach that allows for experimentation and creativity within a stable foundational structure.
Start Small: The philosophy of beginning projects with small, achievable experiments to build momentum.
Questioning Assumptions: The importance of regularly challenging the fundamental reasons and necessity behind building products or processes.

AI Productivity and Block's Experience

The discussion begins by addressing the debate around AI's impact on productivity. While some are skeptical, Block has observed significant gains. Engineering teams that are "AI forward" report saving 8 to 10 hours per week. Across the entire company, Block estimates a 20-25% saving in manual hours due to AI. This metric is expected to increase as AI capabilities evolve.

A key takeaway is that "this is the worst it will ever be" for AI's current impact; it's the baseline, and value is constantly changing, requiring adaptability.

Goose: Block's AI Agent

Goose is described as a general-purpose AI agent, a downloadable desktop tool that acts like a chatbot. It can perform a wide range of tasks, from organizing photos to writing software. Its functionality is enabled by the Model Context Protocol (MCP), an open protocol developed by Anthropic, to which Block was an early contributor. MCPs wrap existing enterprise tools (like Salesforce, Snowflake, SQL) and expose them to LLMs for manipulation, giving AI "arms and legs" to act in the digital world.

Goose is entirely open-source, allowing anyone to download, extend, and write their own MCPs. It can utilize any LLM, including cloud-based models (Anthropic, OpenAI) or open-source models via tools like Ollama.

Examples of Goose's capabilities:

Generating marketing reports by writing SQL to pull data from Snowflake, performing analysis in CSV (using Python), generating graphs with JavaScript, and compiling it into a PDF or Google Doc, then emailing it.
Automating UI tests for mobile applications using a tool called Gling, which acts as a "Goose for mobile" by operating the Android OS natively via accessibility APIs.
An engineer on the Goose team uses it to "watch" his screen and listen to his conversations, proactively building features discussed on Slack or email and opening PRs on Git. It can also nudge him out of meetings if he's running late. This engineer's setup is an experiment pushing the boundaries of AI integration.

Block encourages employees to use any tools they prefer, but Goose is highly integrated due to its MCP foundation, allowing for rapid orchestration of AI with existing systems. Goose can even write its own MCPs, making it bootstrappable. Many companies, including competitors and partners, are actively using Goose.

Organizational Transformation at Block

Dani Prasana's journey to CTO involved a significant organizational shift. He wrote an "AI manifesto" to Jack Dorsey, advocating for Block to become more AI-native.

Key Changes Implemented:

Shifting Identity to a Technology Company: Prasana aimed to re-establish Block's identity as a technology company, similar to Google or Facebook, rather than solely a financial services or fintech company. This involved instituting programs focused on technological advancement, bringing top individual contributors (ICs) together, starting special projects, and reinstituting company-wide hack weeks.
Transition from GM Structure to Functional Organization: This was a crucial step for deeper technological integration.
- GM Structure (Previous): Business units like Square, Cash App, and Afterpay operated as semi-independent companies with their own engineering and design teams, reporting to Jack Dorsey. This was effective for growth stages but hindered deep technological focus.
- Functional Structure (Current): All engineers report to a single head of engineering, and all designers to a single head of design. This allows for a singular focus on driving AI, platform development, and overall technical depth across the company. This mirrors Steve Jobs' reorganization of Apple.

Impact of the Functional Structure:

Unified Technical Strategy: Teams now speak the same language, have access to the same tools, and share common policies.
Talent Mobility: Engineers can move between teams more easily, addressing areas of need.
Focus on Technical Excellence: The organization is now optimized for advancing technical excellence as a primary goal.

How Engineers Work Today vs. Two Years Ago

AI-Native Teams: These teams use "vibe coding" tools and build without writing extensive lines of code manually.
Legacy Codebase Teams: These teams encounter "background AI processes" that run 24/7 or in CI pipelines, analyzing vulnerabilities and attempting to build patches for reported bugs.

Measuring AI Impact

Block measures AI impact primarily through human manual hours saved. This metric is applied across all teams (support, legal, risk, engineering). On average, it's estimated that a fourth of an engineer's time is saved by AI tooling. For engineering teams, the gains are variable, with "greenfield" codebases or new platform apps seeing more aggressive gains than complex, existing codebases.

Future of Engineering and Product Teams (Next Couple of Years)

Increased Autonomy for AI: LLMs will move beyond short, ping-pong-like interactions to work for hours at a time, operating overnight and on weekends to build in anticipation of human needs.
Describing Multiple Experiments: Teams will be able to describe numerous detailed experiments, which AI will build overnight, allowing humans to select the best ones.
Rewriting Over Refactoring: Instead of refactoring existing applications, teams may opt to rewrite them from scratch, leveraging AI to incorporate learnings and improvements efficiently. The goal is to reach a state where a "release" could involve rebuilding an app from scratch.
Human Taste and Judgment: While AI will become more autonomous, human "taste" will be crucial to anchor AI, ensuring it remains tasteful, useful, and valuable, preventing "AI slop."

Building vs. Buying Software

Block prioritizes building its own tools when they serve its core purpose of economic empowerment. While buying SaaS applications might seem cost-effective, it can divert mental bandwidth and technical focus from the company's core mission. The focus should remain on what matters to the company and its customers. Building in-house requires significant maintenance, so the decision should be based on whether the tool directly serves the core purpose.

Hiring and Skillsets in the AI Era

Headcount Planning: The shift to a functional structure has changed how headcount is viewed, moving away from seeing engineers as a commodity to be added for more features. The focus is now on leveraging common platforms, modules, and expertise.
Hiring Criteria: Block looks for individuals with a learning mindset who are eager to embrace and learn AI tools. They are not necessarily looking for expert AI practitioners from the outset. This applies to college graduates and experienced professionals alike.
Interview Process: Interviews are beginning to incorporate AI tools, asking candidates how they would use vibe coding tools to build something and how they think about evolving with them. However, critical thinking and deep technical understanding remain paramount.

Engineer Level Benefiting Most from AI Tools

Senior and Junior Engineers: Both groups are adopting AI tools readily. Senior engineers appreciate the automation of repetitive tasks they've done countless times. Junior engineers are "blitzing through" tasks with these tools.
Non-Technical People: The most surprising and impactful beneficiaries are non-technical individuals using AI agents and programming tools to build software for themselves. This blurs the lines between roles and empowers individuals to optimize their specific workdays.

The Importance of Structure Over Tools

While AI tools are powerful, organizational structure can be more impactful. Block's move to a functional structure, for example, has been a significant driver of efficiency. Even with AI tools like Goose improving build times by analyzing test suites, other factors like offloading tests to the cloud or deleting unnecessary tests can yield greater savings. The question of "do we even need to do this process at all?" is more fundamental than optimizing an existing process with AI.

Core Leadership Lessons

Power of Conway's Law: Understanding how organizational structure directly impacts the systems built is crucial. Changing outcomes requires changing relationships within the organization.
Embrace the Eerie Silence: When things are going well, there's often silence. Leaders need to make time for holistic judgment and reflection, rather than only reacting to problems.
Start Small: Avoid trying to "boil the ocean." Focus on achievable, small experiments that can build momentum. Examples include Goose, Cash App, and Block's first Bitcoin product, all of which started as small initiatives.
Question Base Assumptions: Regularly challenge the fundamental purpose and necessity of building products or processes. This can lead to discovering better alternatives or realizing that a task isn't needed at all.
Focus on Core Competencies: Companies should focus on their core purpose and buy or leverage external solutions for everything else.

Fail Corner: Products That Didn't Work Out

Dani Prasana has a career marked by product failures, including:

Google Wave: An ambitious project that tried to be everything to everyone, starting big with many engineers but lacking user adoption.
Google+: Another significant failure in the social networking space.
Secret: A social networking startup that burned brightly but ultimately failed.
An email startup: Promising but ultimately fizzled out.

These failures provided valuable lessons, fostering humility and a willingness to listen to others' perspectives. Cash App is highlighted as a major success that grew from early learnings from these failures.

Key Takeaways and Advice

Use the Tools Yourself: Leaders and teams should actively use AI tools to understand their strengths, weaknesses, and ergonomics. This firsthand experience is more valuable than reading theoretical articles.
Solve Real Problems: Motivate AI adoption by using the tools to solve specific, personal tasks.
Demand More Openness: In an era of AI, companies should demand more openness from their teams and employers, advocating for open-source contributions and universal benefit, mirroring the original promise of the internet.
Focus on What Matters: Prioritize what is important to you and your company's purpose. Technology should serve these core values, not dictate them.
Don't Fear Change: If you're not energized by your work, make a change. Looking back, monumental problems often seem trivial in hindsight.

Lightning Round Highlights

Recommended Books: Classics, poetry, philosophy, and history (e.g., The Master and Margarita by Mikhail Bulgakov, Tennyson's poetry) over work-related self-help.
Favorite Recent TV/Movie: Alien Earth (pulp sci-fi) and Slow Horses (spy thriller).
Favorite Product: Steam Deck OLED (for its extensibility and customization, challenging the trend of locked-down user experiences).
Life Motto: "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."
Favorite Mad Scientist: Doc Brown from Back to the Future.

Conclusion

Block, under Dani Prasana's leadership, is a prime example of an AI-native company that has strategically transformed its organization and embraced AI tools like Goose to drive significant productivity gains. The emphasis is on adaptability, leveraging open-source principles, and focusing on core company purpose, all while acknowledging that the journey with AI is just beginning. The conversation underscores that while AI is a powerful tool, human judgment, organizational structure, and a focus on solving real problems remain paramount for success.

How Block is becoming the most AI-native enterprise in the world | Dhanji R. Prasanna