Think about ordering pizza. Instead of opening DoorDash, browsing restaurants, customizing your order, and checking out, you could just say "Hey AI, order me a pepperoni pizza." Done. When you think about it that way, why would anyone need DoorDash anymore?

For simple stuff like this, yeah, AI replacing apps feels inevitable. But I've been thinking about the harder cases, and I'm not so sure.

## Most real-world apps solve really messy problems

Take something like Jira. It's not just a pretty interface sitting on top of simple logic. Jira is enforcing workflows across teams, managing who can see what, tracking state changes, hooking into CI/CD, keeping audit trails, all while dozens of people with different roles are working in it at the same time. Can you honestly replace all of that with "Hey AI, manage my project"?

## And then there's the cost

Here's the part that I think people aren't talking about enough.

Let's say you need to build a complex financial model in a spreadsheet. Hundreds of rows, nested formulas, pivot tables, cross-sheet references, the whole thing. Now imagine trying to do that entirely by talking to AI.

"Move that column." "No, the other one." "Now add a VLOOKUP that pulls from the other sheet." "Actually, the range is wrong." Every single back-and-forth is burning tokens. If the model is complex enough, you could easily spend more on AI inference in one sitting than you'd pay for Excel for an entire year. And the spreadsheet app just... does it. Deterministic logic, minimal compute, instant feedback.

I think this pattern holds more broadly than people realize: the more complex and repetitive the task, the more tokens you burn, and the harder it is for "just ask AI" to compete with a $10/month app on cost alone.

Here's where it gets ironic, though. AI coding agents like Cursor and Claude Code are making it way cheaper to build apps. So AI might not shrink the app market at all. It could actually grow it by making apps cheaper to create, while simultaneously making "just use AI directly" expensive for anything non-trivial.

*Curious what HN thinks:*

- Are we overestimating AI's ability to replace purpose-built software?

- Will inference costs drop and AI capabilities advance enough to make this argument irrelevant?

- Or will everyone just become a developer, building their own apps for their own needs?