You can build anything now. That's exactly why you shouldn't build this.

Every software company is hearing the same question now: with AI, why buy anything we could just build ourselves? It's a fair question, and half of it is true. Most of it really is easy to stand up now: describe what you want and an agent writes the first version before lunch. You could build the whole thing.

The honest answer starts somewhere uncomfortable. You're right. And it doesn't matter.

The middle is disappearing

The people who fund and run software for a living have mostly worked out where this lands. Two of them draw the same map.

Microsoft CEO Satya Nadella argues the classic business app is about to collapse. Most of it was only a "CRUD database with a bunch of business logic", and in the agent era that logic moves up into an orchestration layer across many systems. So the app-shaped layer you'd build, the screens and glue around your data, has the shortest shelf life left. What survives is the data underneath.

Seema Amble at a16z gives the split a name, software losing its head. The head is the interface, the part humans used to click through, and it commoditizes now that agents do the clicking. The body is everything under it, and it's where the value pools, because "agents may kill muscle memory as a moat, but they do not kill operational logic and context." The head is the dashboard; the body is the permissions model, the audit trail, the retries and failovers, the compliance nobody notices until it's missing. Teams reach for the head. The body is the part that's actually hard, and the part that matters.

So where does the value go? To the two ends: the infrastructure everything runs on, and the product you build on top of it. What sits between, the screens and glue, is what you now have AI generate on request, easy to build, easy to copy, worth less every month. Which splits your job in two. One half is the product only you can build, and the users you obsess over to keep. Build that. The other half is the infrastructure that just has to work. Buy it, and operate it.

Don't build it yourself

Not because you can't. Because the moment you build it, you own something that was never your product, and owning it is the part that doesn't get cheaper.

Building is cheap. Everything after it is not. What you make is now yours to secure, patch, and answer for at 3am, for as long as it lives. Code is a liability, not an asset: the capability is what you wanted, the code is the standing cost of having it. AI doesn't change that. It generates the liability faster, and it's weakest exactly where the cost is highest: the big, stateful, tangled systems, and the operating that never stops. The more you build, the more you own.

Ownership is really a claim on attention, the one thing AI doesn't make more of. It hands you infinite ability to build and not one extra hour to mind what you've built. When anyone can build anything, building stops being the scarce thing. Attention becomes the whole game, and the only question that matters is where you spend it. The infrastructure underneath should cost you nothing to think about, the way you notice electricity only when it's out. Everything you run yourself is a standing draw on the one budget that never grows, one more thing your team holds in its head instead of the product that's actually yours.

Let an agent run it

The move isn't to build more because building got cheap. It's to build less. Building it yourself costs more than the plumbing; it costs the focus you'd have spent on your own product. So give the AI the other job: not building your infrastructure, but operating it.

Build-vs-buy used to be two choices. It's three now: build it, buy it, or buy something opinionated and let an agent run it. That third option is the one that wins: the agent configures it, wires it into your product, and reads back what happened, with no human toil and no plumbing to own.

Even at a fair price, the return isn't the line item. It's that no one on your team thinks about it again, and that attention goes back to the product only you can build.

Will you ever get to stop?

So, back to the objection. You can build any of it yourselves, and with an agent you can build it fast. That was never in question. The question is whether you ever get to stop.

You can build it. The real win is never having to.