For two years, developers treated GitHub Copilot like an all-you-can-eat buffet. Ten pounds a month bought unlimited code suggestions, unlimited chat queries, unlimited refactoring passes. That deal just ended.
This week, GitHub switched Copilot to token-based pricing - and developers are furious. Not because the pricing itself is outrageous, but because the change kills the behaviour that made Copilot useful in the first place.
The Psychology of Unlimited
Under the old model, you could hammer Copilot with half-formed questions, iterate on solutions, ask it to explain unfamiliar code twenty different ways. There was no mental friction. No metre running. You used it like you'd use a colleague - freely, messily, iteratively.
Token pricing changes that calculation completely. Now every query has a cost. Not a huge cost - but enough that you start second-guessing. Is this question worth it? Should I try Google first? Maybe I can figure this out myself.
The moment you introduce that friction, the tool becomes less useful. Not because it works differently, but because you use it differently. You ration. You hesitate. The value drops even if the price stays reasonable.
Enterprise Tightens the Screws
For individual developers paying their own subscription, this is annoying. For developers inside companies, it's worse.
Engineering managers are already getting emails from finance asking why Copilot costs spiked 40% last month. Teams are being told to "use it more efficiently" - which translates to "use it less". Some companies are setting monthly token caps per developer. Others are restricting access to senior engineers only.
The developers who benefited most from Copilot were junior engineers learning unfamiliar codebases. They're the first to lose access when budgets tighten. The seniors who need it least get to keep it.
The Bigger Pattern
This isn't really about GitHub. It's about the end of the AI honeymoon period.
For the last eighteen months, AI companies subsidised usage to build habits. OpenAI, Anthropic, GitHub - they all ran their services at a loss to get developers hooked. Now the bill is coming due. Prices are going up. Free tiers are shrinking. Usage caps are appearing.
The shift from flat-rate to consumption pricing is spreading across the industry. ChatGPT's new professional tier? Token-based. Claude's API? Token-based. Gemini? Same story. The unlimited era is over.
What It Means for Builders
If you're building on top of these platforms, this changes your economics overnight. The cost of running an AI feature used to be predictable - now it scales with usage in ways that can blow up your margins.
A chatbot that used to cost fifty quid a month might suddenly cost five hundred if your users get chatty. A code analysis tool that was profitable at scale might not be anymore. You need to build token budgets into your product design from day one.
For individual developers, the calculation is simpler but no less painful. You're about to get a lot more selective about when you reach for AI assistance. That might make you a better programmer - forced to think through problems yourself - or it might just slow you down. Probably both.
Either way, the era of treating AI like a bottomless resource just ended. We're back to thinking about costs. Which means we're back to thinking about value.