
"We're Not Writing Code by Hand Anymore. That's Over." | Owen Jennings & David Haber - The a16z Show
AI Summary
Block recently made a significant decision to reduce its workforce by over 40%, a move driven primarily by the transformative impact of AI on productivity and operational efficiency. This decision was rooted in a multi-year effort to integrate agentic development, which began with the launch of Goose, an internal agent harness, in early 2024. While initial progress in augmenting software development and internal tooling was steady through 2024 and 2025, a binary shift occurred in late November/early December with the emergence of advanced foundational models like Opus 46 and Codex 53. These models proved incredibly capable of working with complex, existing codebases, fundamentally altering the correlation between the number of employees and company output.
Historically, the number of people at a company directly correlated with its output. However, Block observed that one or two engineers or a designer and an engineer using these new AI tools could be 10, 20, or even 100 times more productive. This paradigm shift led Block's executive team, including CEO Jack Dorsey, to spend Q1 deliberating on the fundamental implications for product development, software building, and company operations. The resulting reduction in force was heavily concentrated on the development side, indicating that it was a response to technological change rather than an overhang from past overhiring. Cuts in areas like outbound sales or account management were minimal.
Block approached this transition from a position of financial strength, meaning the layoffs were not driven by financial distress but by a strategic vision for the optimal organizational structure in an AI-driven world. Core principles guided the process: ensuring reliability to prevent outages, building trust with customers and maintaining compliance in a complex regulatory environment (e.g., the compliance team was largely untouched), and continuing to drive durable growth by building new features and making long-term bets, albeit with smaller teams. The organization was rebuilt from scratch, with development teams looking completely different, while other areas like regulatory counsel remained similar.
The execution of the reduction in force was handled deliberately and generously, acknowledging the long-standing relationships with many departing employees. Severance packages were generous, technology access was not immediately cut, and an all-hands meeting with Jack Dorsey and the executive team explained the decision and its drivers to the entire company. Following the initial shock, Block implemented significant operational changes, including an 70-80% reduction in meetings, allowing employees more time for actual building and work. Weekly all-hands meetings with Jack foster continuous communication and alignment.
The most meaningful difference in operations post-reduction is the shift from a linear workflow to one where individuals manage multiple AI agents working in the background. For example, a project that once required 15 people now might be handled by four people plus AI tools with unlimited access to tokens. This forces a change in workflow, where individuals context-switch between agents building pull requests (PRs) on their behalf, rather than sequentially submitting and reviewing PRs. This applies not only to software development but also to product managers and growth marketers, with individuals now overseeing numerous agents.
Block's internal AI infrastructure includes "Goose," an agent harness that is model-agnostic, capable of running on various large language models (LLMs). This platform forms the foundation for many automations within Block. An internal agentic operating system called "G2" allows anyone to automate deterministic workflows. On the development side, tools like "Builderbot" autonomously merge PRs and build features, often completing 85-90% of complex features before human refinement. This has massively compressed the time from idea to product launch. Beyond development, AI is automating deterministic workflows across operations, customer support, risk, and compliance, with chatbots and AI phone support handling a majority of inquiries. While a human-in-the-loop approach is currently critical, the long-term vision is that these AI systems will outperform humans in many decision-making tasks.
Block's business structure has also evolved. Formerly operating with separate business units for Square and Cash App, the company functionalized about 18 months ago, with all engineering, design, and product teams reporting to centralized heads. This fosters the creation of a unified financial platform and a business platform that spans the entire company, building features and products that connect Square, Cash App, and Afterpay. Cash App, in particular, has seen significant growth, now accounting for approximately 60% of Block's overall gross profit.
On the product side, the biggest shift is towards "generative UI," moving away from static, rigid user interfaces. In practice, this means personalized app experiences where the UI dynamically generates based on user preferences and behavior. For example, Moneybot within Cash App can generate charts and visualizations on the fly in response to user queries about spending habits. Similarly, ManagerBot on the Square side can create custom apps for business owners, like a scheduling app for a multi-location restaurant, with the app's look and feel not hardcoded into the application's source. This offers greater control, higher engagement, and better product experiences. Block emphasizes proactive intelligence, where the system prompts customers with relevant information and actions, especially concerning financial management.
Regarding defensibility, Block identifies several moats in the near to medium term, including distribution and network effects (e.g., 50-60 million monthly active users), licenses and regulatory posture, and hardware that is currently difficult to "vibe code." However, in the longer term, the key to defensibility will be a company's deep understanding of something inherently difficult for others to grasp. Block aims to evolve into an "intelligent system" that continually iterates and improves its understanding of how sellers and buyers participate in the economy. This involves building internal and external "world models" of customers and Block's own operations. The ultimate vision is a loop where a company's deep insights ("signal") are fed into agentic tools (like Builderbot or Claude Code), allowing for rapid iteration and building. This loop, which currently takes weeks with human involvement, is projected to run hundreds or thousands of times a day in the future, with humans potentially transitioning to more editorial roles. The companies that master this continuous, rapid iteration based on unique, hard-to-understand insights will possess the strongest moats.