Remember when an Uber across town was $7? That price was a subsidy, meant to get you hooked on the convenience until taking a taxi felt unthinkable. Once that happened, prices normalized, and now the same ride is $35.

I think AI tools are running the same play.

Right now, companies are getting huge credits to put their engineering teams on Copilot, Cursor, Claude, and a dozen other tools. The pitch is always the same: look how fast your team moves, look at the productivity gains, and it costs almost nothing, so sign everyone up and let them build their workflows around it.

The price goes up once your team can’t ship without an AI assistant, your processes depend on it, and your engineers have lost some of the muscle for writing code from scratch. At that point you aren’t leaving, because the switching cost is too high and you really do depend on it.

You can already see it starting. GitHub Actions recently reworked its pricing in ways that shift costs up for teams that use it heavily. Anthropic’s plans are usage-based, with hourly and weekly caps that hit hardest on the exact workflows they encouraged you to build. Free tiers keep getting tighter, pro tiers keep getting more expensive, and the enterprise tier requires a conversation with sales.

I don’t think anyone is plotting this. It is just how platforms make money: subsidize adoption, get users dependent, then charge for the dependency.

So what do you do about it?

I am not telling anyone to avoid AI tools. That would be like telling someone in 2015 to just take the bus. The tools are useful and the productivity gains are real, but build with the expectation that what is nearly free today won’t stay that way.

Design so you can swap tools out, and don’t let a single vendor’s tool become the only way your team can work. Keep your engineers sharp enough to work without the assistant, and take stock of what you actually depend on and what it will cost once the credits run out.