How To Build A Company With AI From The Ground Up
Y Combinator Partner Diana Hu explains how to build an AI-native company where AI is the core operating system, not just a tool. She covers how to make a company queryable, the impact on team structures, and why startups have a massive edge in this new paradigm.
The fundamental shift in company building is not about using AI for incremental productivity gains, but about rebuilding the company to run on AI as its core operating system. Every workflow, decision, and process should flow through an intelligent layer that is constantly learning and improving. This approach moves beyond simply adding copilots to existing workflows and instead focuses on unlocking entirely new capabilities.
From Open-Loop to Closed-Loop Systems
Historically, companies have operated as open-loop systems: decisions were made and executed, but outcomes were not always systematically measured to adjust the process. This method is inherently lossy.
The AI-native approach is to operate as a closed-loop system. Such a system is self-regulating; it continuously monitors its output and adjusts its processes to better meet its stated goals. By implementing self-improving agents, a company can run as a powerful, stable, and correct closed-loop.
Building a Queryable Organization
To enable these closed-loops, the entire company must be made queryable and legible to AI. Every significant action should produce an artifact that the central intelligence can learn from. Practical steps to achieve this include:
- Recording meetings with AI notetakers.
- Minimizing private communication like DMs and emails in favor of open channels.
- Embedding agents throughout all communication channels.
- Building custom, comprehensive dashboards that track everything from revenue and sales to engineering and operations.
A concrete example is engineering management. An agent with access to Linear tickets, Slack channels, customer feedback (e.g., from Pylon), high-level plans in Notion, sales calls, and standup recordings can analyze what was shipped and how well it met customer needs. Going further, this agent can propose more predictable and accurate sprint plans, eliminating the need for lossy status rollups from engineering managers. Teams adopting this have seen sprint times cut in half while achieving nearly 10x more output.
The overarching principle here is that to get their full capabilities, you need to provide models with as much context as you would provide an employee.
The Rise of AI Software Factories
A new paradigm emerging for high-velocity product development is the AI software factory, an evolution of test-driven development (TDD). The workflow is:
- A human writes a detailed specification and a set of tests that define success.
- AI agents generate the implementation and code.
- The agents iterate on the code until all tests pass.
The human defines what to build and judges the final output, while the agent handles the how. Some companies have already pushed this to the point where their repositories contain no handwritten code, only specs and test harnesses. This is the path to creating the 1,000x or 10,000x engineer: a single individual surrounded by a system of agents, enabling them to build what was previously impossible.
The New Organizational Chart
This new way of building renders the classic management hierarchy obsolete. Middle managers, who traditionally routed information up and down the organization, are replaced by the central intelligence layer. Since a company’s velocity is dictated by its information flow, removing human middleware provides a direct speed gain.
Drawing on Jack Dorsey’s thinking at Block, the company itself must be rebuilt as an intelligence layer with humans at the edge guiding it. This leads to three new employee archetypes:
- The Individual Contributor (IC): The builder and operator. In an AI-native company, everyone—including support and sales—is a builder, coming to meetings with working prototypes, not just presentations.
- The Directly Responsible Individual (DRI): A strategist focused on customer outcomes, with clear, direct responsibility for a result. This is not a classic manager.
- The AI Founder: This person still builds and coaches, leading by example. The founder must be at the forefront, demonstrating the massive capability gains possible with AI, not delegating the strategy.
Maximizing Tokens, Not Headcount
The critical metric for success will shift from headcount to token usage. The best companies will be “token maxing,” willingly running high API bills because it replaces what would have required a far more expensive and inflated headcount. One person with the right AI tools can now match the output of a large pre-AI team.
Founders cannot outsource their conviction on this. They must personally use these tools to break their own priors about what is now possible.
The Startup Advantage
Early-stage founders have a monumental advantage in this shift. They are not burdened by legacy systems, entrenched org charts, or the need to retrain thousands of employees. They can design their systems, workflows, and culture around AI from day one.
Incumbents face the challenge of unwinding years of standard operating procedures while maintaining a live product. While some may succeed by creating isolated “skunkworks” teams, most will find it difficult to change their core processes. Startups have no such constraints, allowing them to operate thousands of times faster and seize this opportunity.