Personal assistants don't have product-market fit...yet

There is a lot of hype right now around personal assistants. Meta has Muse, Google has Gemini Spark, Apple has Siri AI, X has Grok Bot, OpenAI has ChatGPT, Ollie is already the default on a phone near you, and Instinct showed up promising to out-think all of them. Each launch comes with the same pitch: you need this. What's less clear, every time, is need it for what, exactly?

Look closely at the use cases people actually cite and a pattern shows up fast. Find the duplicate charge on last month's statement. Call the internet provider and negotiate the bill down. Cancel the subscription you forgot you had. These are real problems and a personal assistant is genuinely good at solving them. But they are also, almost by definition, things that happen once. You find the duplicate charge, you get the refund, and the problem is gone. You renegotiate the bill, and that specific bill stays renegotiated for a year. What is the assistant supposed to do with itself the day after?

A signal, not a product yet

I don't think the hype is wrong about where things are headed. Handing a piece of software enough context and enough permission to act on your behalf, not just answer questions about your behalf, is a real shift, and probably the more important one this wave of AI has produced. But being right about the direction of a technology and being ready for people to build a daily habit around it are two different things, and most of this round of launches is confusing the first for the second.

This isn't a new mistake. Plenty of technologies that were directionally correct still needed a second or third run at the market before the habit actually formed. The pattern is familiar: the first wave over-promises on the everyday use case, the market gets excited, tries it, doesn't find a reason to open the app the next morning, and quietly drifts off. The tech was not wrong. The timing was.

The part that should worry us more than it does

Here is the part of the cycle that's genuinely different this time, and worse. A personal assistant that's actually useful needs broad access to be useful at all: your inbox, your calendar, your accounts, your messages. That's not a nice-to-have permission, it's the entire premise of the product. So we grant it, because the one-off task in front of us is worth solving.

Then the one-off task gets solved, and there is no second task queued up behind it, because the everyday use case was never really there. We stop opening the app. The permissions don't revoke themselves. Somewhere between the excitement and the drift, there is a window where a piece of software neither of us is thinking about anymore is still sitting on top of our entire digital life, and the first time most people find out is when something goes wrong.

What day-to-day actually requires

Every durable product that lives on a home screen has an answer to a boring question: what does this do for me on a Tuesday when nothing unusual is happening? Search has an answer. Messaging has an answer. A calendar has an answer. Right now, most personal assistants only have a good answer for the Tuesday when something unusual is happening, the duplicate charge, the bill, the cancellation. The rest of the week, there's nothing pulling you back.

That's a product-market fit problem, not a capability problem. The model is capable enough for a lot of this already. What's missing is a reason to come back that doesn't depend on something going wrong in your finances first.

Bottom lines

A real problem solved once is not a habit. The current crop of assistants is strong at rare, high-value, one-off tasks, and weak at giving people a reason to return the next day.

Broad permissions granted for a narrow job are a standing liability. The access doesn't shrink back down when the task is done, and most people never think to check.

Hype arriving early doesn't mean the thesis is wrong. It usually means the market needs a round or two to mature before the habit-forming version shows up.

Watch for the second introduction. Six, maybe twelve months out, as the tech and the surrounding trust infrastructure catch up, we'll be introduced to this category again, and that's likely the version that sticks.

None of this means personal assistants are a dead end. It means we're early enough in the cycle that the pitch outruns the daily use case, and the gap between the two is exactly where the permissions problem lives. The category will get its real product-market fit. It's just probably not this round.