What an AI-native company actually looks like

I watched a talk from Y Combinator recently that put words to something I’ve been living for the past year. Credit where it’s due: Diana Hu’s talk on AI-native companies sparked this article.

Her talk is aimed at software startups. The ideas apply to any business. A lender, an agency, a housing provider, a coaching company. All of it translates. Here’s what it means for the rest of us.

The wrong question

Most businesses are asking “how can AI make my team more productive?”

Buy some licences, add a copilot, make everyone 10 to 20 percent faster at the job they already do. That’s the frame almost everyone is using, and it misses what’s actually happening.

The real shift is capability, not productivity. The right person with the right AI setup can now do work that used to take a whole team. Not slightly faster. Categorically different.

Once you see it that way, the question changes. It’s not “which AI tools should we buy?” It’s “how should the business itself be built?”

AI as the operating system

AI shouldn’t be a tool your company uses. It should be the operating system your company runs on. Every workflow, every decision, every process flows through an intelligent layer that keeps learning.

The best way I’ve heard it framed is open loops versus closed loops.

Most businesses run as open loops. You make a decision, execute it, move on. The outcome rarely gets measured, and the lesson rarely makes it back into the process. Knowledge lives in inboxes and people’s heads. Status comes from asking around. Every lesson gets learned again next quarter.

A closed loop is different. It captures what happened, feeds it back, and adjusts. It self-corrects. Run your important processes as closed loops with AI in the middle and the improvement compounds every week.

Making your business queryable

None of that works if AI can’t see your business. The whole company has to be queryable. Every important action should leave a record an intelligent system can read and learn from.

In practice:

• Record your meetings with an AI notetaker. Decisions and commitments become searchable, not remembered.
• Get real work out of DMs and email. Run it through shared systems where progress is visible.
• Put your numbers in dashboards. Revenue, sales, delivery, ops. One live view.
• Give AI the same context you’d give a trusted employee. That’s the standard.

I can vouch for this part personally. I run Axela this way. Every client meeting is transcribed and searchable. Client status, proposals, delivery notes and finances live in one place an AI can read. When I ask “what’s the state of this account?” I get an answer built from everything that actually happened, not from what I remember.

The honest bit: it took months of habit change, and the first few weeks feel like admin. Then it compounds, and you stop chasing status entirely.

What happens to the org chart

Here’s the part of the talk I found most confronting.

Middle management exists mostly to route information. Collect updates from below, summarise for above, coordinate sideways. Once the business is queryable, an intelligent layer does that routing instantly and without the loss.

Jack Dorsey has landed on the same conclusion at Block. His model: three roles going forward.

1. Builders. People who directly make and run things. Not just technical people. Ops, sales, support, finance. Everyone builds, and everyone brings working prototypes to meetings instead of slide decks.
2. Owners. One person, clearly responsible for one outcome. No hiding behind committees.
3. AI-first leaders. Leaders who still build and lead by example. If you run the company, this one is yours. You can’t delegate your AI conviction to a committee or a consultant.

The line from the talk that stuck with me: you should be willing to run an uncomfortably high AI bill, because it’s replacing what would have been a far more expensive headcount. The best companies will maximise AI usage, not headcount.

Why smaller businesses win this one

This is the part I’d underline for every mid-market leader in Australia.

Large enterprises have years of standard operating procedure to unwind, entrenched org charts, and thousands of people to retrain. Every change to a core process risks breaking something that already works. Most of them will manage a pilot program and a policy document.

A 20 to 200 person business doesn’t have that problem. You’re small enough to redesign around AI now. Loops and dashboards can be live in weeks. Culture gets set from the top, by example.

The advantage isn’t budget. It’s structural. And it’s one of the few times the smaller player has the edge over the incumbent.

Where to start

Not with a tool purchase. But not with a scatter of AI pilots either.

The trap on the other side is picking a clever use case, bolting an agent onto it, and calling it progress. It demos well, then dies, because there is nothing underneath it. No shared knowledge for the agent to draw on, no connection into the systems where work actually happens, no guardrails that let it act safely. The pilot becomes a screenshot in a deck.

Start with the foundation, built small:

• The brain. One queryable place for what the business knows. Meetings, decisions, client history, numbers.
• The sockets. The connections that plug intelligence into the systems where work actually happens, so it can act, not just answer.
• The harness. The checks, guardrails and human sign-offs that let agents do real work without breaking things.

Then close your first loop on top of it. Pick one process that matters, make it visible, let the system learn from it. The first loop proves the value. The foundation is what makes the second, fifth and tenth loop cheap. That is the part that compounds.

And the point I’d act on first: you can’t outsource conviction. Sit with these tools yourself until they break your assumptions about what’s possible. Mine broke months ago and I haven’t looked back.

If you want to see what this looks like in your business, that’s the work we do.