The great unbundling

A meeting-notes app just raised money at a billion and a half dollars. Not a platform. Not an operating system for work. An app that listens to your calls and writes down what was said. If that sentence makes you wince, you’re having the right reaction — and the wrong conclusion.

The wince says: that’s a feature, not a company. And you’re right, in the old grammar of software. But the old grammar is breaking, and the break is the most interesting thing happening in products right now. It’s worth understanding before it reprices your whole mental model of what a product even is.

The bundle was a technology tax

For thirty years, software wanted to be a suite. The logic was iron: distribution was expensive, switching costs were your moat, and every feature you added made the customer a little more trapped. The suite wasn’t a gift to the user. It was a tax they paid because the alternative — stitching ten tools together yourself — was worse.

Bundling solved real problems: one login, one bill, one vendor to blame, data that moved between features without a fight. But it also carried a lie. You were told the integrated suite was better, when often it was just cheaper than the friction of unbundling. The all-in-one was never the best notes app or the best design tool. It was the most tolerable compromise.

AI is quietly removing the reason the compromise existed.

Why the single feature suddenly stands alone

Two things changed at once.

First, building a genuinely good single-purpose product got radically cheaper and faster. A small team with AI in the loop can now ship something polished that would have needed a department five years ago. The floor on quality rose while the cost of reaching it fell. That’s why the “that’s just a feature” products keep winning — the feature, done obsessively well, is now a viable company.

Second, and more importantly, the integration problem is dissolving. The reason you tolerated the mediocre bundled tool was that your data lived there and moving it was painful. But when a model can read from anywhere, act across surfaces, and carry context between tools, the lock-in that made bundles defensible starts to leak. Standards for tools to talk to each other — the plumbing that lets one product feed another — are turning “my data is trapped in the suite” into a solvable problem.

When integration is cheap, best-of-breed beats all-in-one. That’s not a prediction. It’s just what happens to a compromise when the thing forcing the compromise goes away.

The trap on the other side

Here’s where I’d push back on my own argument, because the unbundling story has a brutal second act that the excited version skips.

If a single feature done well is now a company, then a single feature done well is also trivially copyable — by the incumbent whose suite you’re unbundling, and by the foundation model whose capabilities you’re thinly wrapping. The same forces that let you build fast let everyone build fast. The moat that used to come from integration doesn’t automatically reappear somewhere else.

Watch what the smart single-feature companies do the moment they get traction: they stop being a single feature. The notes app rebrands as an “enterprise context layer.” The search box becomes a browser and an agent. They race to build the one thing that AI doesn’t commoditize — accumulated context, proprietary data, workflow lock-in, a brand people trust — before the copy arrives. Unbundling is how you get in. It is not how you stay.

So what is a product now?

This is the question I keep circling, and I don’t think anyone has a clean answer yet. But here’s my working one.

A product used to be a bundle of features you owned. Increasingly, a product is a point of view about a workflow, wrapped around a defensible sliver of context or trust, with the commodity capabilities rented from a model underneath. The features are table stakes. The model is a utility anyone can call. What’s left — the actual product — is the judgement about what to build, for whom, and the accumulated context that makes it stickier than the copy.

That’s a strange thing to sell, and it’s why marketing these products is getting harder. You can’t win on the feature list anymore, because the feature list is copyable by Friday. You win on being the sharpest possible answer to a specific question the customer has — and on being trusted enough that they don’t go shopping when the copy appears.

The billion-dollar notes app isn’t absurd. It’s a signal. It says the suite’s protection is gone, that a single obsession can be a business, and that the whole map of who-owns-what in software is being redrawn in real time. The wince you felt is your old model protesting. Update the model. The unbundling is real — and so is the scramble to re-bundle around something that AI can’t simply reproduce.