The tyranny of software really is almost over. That part is true. What’s getting missed is the new one we’re walking into.

The Verge published a piece today arguing that AI-assisted development is putting software creation in the hands of ordinary people: lawyers who want case management that actually fits their practice, doctors who want workflows built for how they work rather than how a product team imagined they work, churches that need something other than a jailbroken spreadsheet. The premise is correct. The framing is too cheerful by half.

The interesting question isn’t whether you can make software now. You can. The interesting question is what happens when the assumption that software is built for the median user stops being true across the whole economy.

The Median User Was Always a Fiction

Enterprise software is priced the way it is because companies sell one product to millions of people. Salesforce charges what it charges because it’s the same Salesforce for a Fortune 500 company as for a ten-person startup. The features are the features. The interface is the interface. You adapt to the software, not the other way around.

This model works because the alternative, custom software, used to cost hundreds of thousands of dollars and a dedicated team to maintain. AI doesn’t eliminate that cost. It makes it cheap enough that it disappears from the conversation as a line item. For a small business owner, that’s a category change.

That’s a bigger deal than the coverage suggests.

What’s Actually Being Disrupted

The SaaS market is enormous, and most of its value is captured not because the software is great but because switching costs are brutal and the product does enough of what you need that you learn to live with the gaps. Every enterprise software sales team knows this. The pitch is always “we’re good enough, and the cost of leaving is high.”

AI-assisted development doesn’t just let individuals tinker. It lets small businesses build workflows that would have previously required a six-figure implementation. It lets internal teams build integrations that used to require a vendor contract and a six-month wait. Anthropic is explicitly targeting the 36 million small businesses in the U.S. right now, which tells you something about where the company thinks the real growth is.

The threat to legacy SaaS isn’t that individuals will replace enterprise software. It’s that the middle market will stop buying it at all, because they can build something better in a weekend that actually fits how they work.

The Part Nobody’s Talking About

Here’s what keeps nagging at me: if everyone’s software is different, what happens to shared conventions?

Right now, software creates shared context. Everyone who uses Slack knows what a channel is. Everyone who uses Notion knows what a database is. That shared vocabulary makes collaboration possible and makes it easier to hire people who can hit the ground running. It’s also how we train people to navigate digital tools in general.

When your legal practice is running on a custom-built AI app your paralegal put together last Tuesday, and my legal practice is running on something completely different, coordination gets harder. Security gets messier, because who’s auditing the vibe-coded billing system? Onboarding new employees means reteaching basic workflows from scratch every time.

These aren’t dealbreakers. They’re real costs, and almost nothing being written about personal software right now accounts for them.

Where This Is Actually Heading

Anthropic’s Cat Wu said yesterday that the next big step for AI is proactivity: AI that anticipates your needs before you know what they are. Google is already rolling out something it calls “contextual suggestions” to Pixel 10 devices, predicting your next action based on location and daily habits. The system notices you’re at the gym and offers to start your usual playlist before you ask.

The end state isn’t “you build your own software.” The end state is software that builds and adapts itself around what it observes about you. Your tools reshape themselves to fit your patterns rather than the other way around. That’s genuinely different from everything that came before it.

But here’s the catch. AI that anticipates your needs is only as good as its model of you, and building that model requires watching you constantly. The vision of perfectly personalized software and the vision of persistent behavioral surveillance are the same vision. You’re going to get both together, not one without the other.

The Actual Deal on the Table

Personal software is real, and it’s going to reshape a significant piece of the software market. The lawyers, doctors, and small business owners who couldn’t afford custom software before are going to get it now. Some of it is going to be genuinely better than what they were using, and that matters.

But the framing of “the tyranny of software is almost over” is too clean. We’re trading the tyranny of the median-user problem for a dependency on whoever controls the AI doing the building and adapting. The software becomes more personal. The infrastructure underneath it becomes more concentrated.

That’s the actual deal. It’s still a good deal for a lot of people. But you should know what you’re signing.