Robert Ta and Jonathan discuss how AI pricing swings are becoming an engineering problem. GitHub moved to token-based billing for Copilot, OpenAI raised GPT-5.5 prices, and Fairmind found costs fluctuating 50-90% in 48 hours. Teams may soon need token budgeting as a standard part of sprint planning, similar to cloud cost management.
We've already seen GitHub moved away from flat rate billing to token-based billing. Saw that recently and they're doubling down on GitHub Copilot and allowing anybody to build their own harness. That's kind of the platform play they're making. OpenAI doubled their GPT-5.5 pricing and then Fairmind noted AI cost volatility of 50 to 90% inside 48 hours is now a real planning risk. That's really interesting is will token budget optimization be a part of sprint planning cycles in the future?
The token budget management that you're bringing up could probably come comes into the, just the maintainability bucket. Is this valuable? Is it feasible? Is it maintainable? You know, just like cloud budgets, I think there'll be a point where it's like a job of every team to make sure that this is a maintainable and sustainable project from a budgeting perspective. It hasn't needed to get there yet, uh, but it, it's quickly pushing that direction. Aren't you curious what's also going to happen? Like Anthropic reported, I think it was like $47 billion in annual revenue. Part of it was this half a billion dollar spend from a company who's not going to spend that much next month. And Part of it's Uber who ran out of budget and so on. What do you think? Do you think we'll see like a little bit of a dip in their trajectory or just kind of like maybe a flattening off or?
Yeah, that's a really good question, Jonathan, which is, are Claude and OpenAI's run rates magnified by poor budget planning and a lack of the industry's understanding of token optimization to then capture economic value from their tokens?
