AI Is Fading Your Company. Here's How to Win.

In Brief

Most companies talk about AI as a productivity booster. Y Combinator Partner Diana Hu says that's the wrong frame entirely. The companies that will survive the next decade won't just use AI. They'll run on it, the way software runs on an operating system. And the startups that build this way from day one have an edge established companies can't easily close.

AI as the operating system

The shift Diana Hu is describing isn't about adding AI to existing workflows. It's about making AI the foundation everything else runs on.

"It should be the operating system your company runs on," she says. "Every workflow, every decision, and every process should flow through an intelligent layer that is constantly learning and improving."

This reframing has a specific implication: AI stops being a tool your employees use and becomes the infrastructure your company runs through. That's a fundamentally different relationship with the technology. The consequence is that many of the tools a company has always relied on gradually lose their role.

The company that learns from itself

Most companies have historically operated as what control systems engineers call an open loop: you make a decision, execute it, and don't systematically measure what happened or adjust the next decision based on results. Information gets lost. Feedback is informal.

A closed loop is self-regulating. It monitors its own output continuously and adjusts the process to better meet the goal. Hu's argument is that every important process in your company should run as a closed loop powered by AI, so the system learns from what actually happened, not just what someone reported upward.

Make everything in your company visible to AI

For closed loops to work, the AI at the center needs context, the same context you'd give a new employee to make them effective. Hu calls this "making your company queryable": structuring the organization so that AI can read and learn from everything happening inside the company.

In practice, that means recording meetings with AI notetakers, moving decisions out of DMs and into trackable channels, embedding AI agents across communication tools, and building dashboards that bring revenue, sales, engineering, and ops data together in one place. For the individual employee, that means instead of opening ten different applications, you talk to one AI interface that handles everything.

The concrete example she gives: an engineering agent with access to Linear tickets, Slack channels, customer feedback, GitHub, and sales call recordings can analyze what actually shipped in a sprint (a fixed work period, typically two weeks) and how well it met real customer needs, then propose the next sprint plan with far more accuracy than any manual status report. Teams that have implemented this have cut sprint planning time in half and gotten close to 10x more done.

Software factories and the 1000x engineer

The next evolution is what Hu calls software factories. The concept borrows from test-driven development (TDD, a coding practice where you write tests before writing code): humans write a specification and a set of tests that define success, then AI agents generate the implementation and iterate until the tests pass. The human defines what to build and judges the output. Writing the actual code is the agent's job.

Some companies have already pushed this to the point where their code repositories contain no handwritten code at all: just specs and test harnesses.

This is the mechanism behind what tech blogger Steve Yegge describes in his essay The Anthropic Hive Mind as the 1000x engineer: a single engineer surrounded by a system of agents that enables them to build things that previously required an entire team. "The era of the 1000x or even 10,000x engineer is here," Hu says.

When middle managers disappear

Build a company this way (AI loops everywhere, a queryable organization, software factories) and the classic management hierarchy stops making sense.

Middle managers existed to route information up and down the organization. That routing function is now the intelligence layer's job. Every layer of human coordination you can remove is a direct speed gain.

Jack Dorsey, co-founder of Twitter and CEO of payments company Block, has been rebuilding it around these principles and describes three roles for the employees who remain in an AI-native company:

1. IC (Individual Contributor): The builder-operator. Not just engineers. Everyone builds: support, sales, ops. Everyone shows up with working prototypes, not pitch decks.
2. DRI (Directly Responsible Individual): Owns a specific outcome. One person, one result. No hiding behind a team.
3. AI founder type: Still builds, still coaches. If you're the founder, this is your role: show your team what the new capabilities look like, not delegate your AI strategy to someone else.

The metric that matters in this world isn't headcount. It's token usage: how much AI compute your company runs through to get things done.

Startups have the edge

Existing companies face a painful problem: they have to unwind years of standard operating procedures and assumptions about how software gets built, while keeping a live product running. Every change to core processes risks breaking something that already works.

Startups don't have that constraint. You can design your systems, workflows, and culture around AI from day one. "You can operate a thousand times faster than the established companies," Hu says.

That's the window. And it's open right now.

Glossary

Sources and resources