It’s already happening.
As declared in my previous article “AI Boosters Are Rude (On AI Hype)”, token cost has begun to noticably surge. Or rather, casual consumers of LLM subscriptions are starting to feel the true cost of tokens for the first time. Behind the scenes, updates have been rolled out to Codex, the terminal agent offering from OpenAI (ChatGPT), substantially reducing its limits for paying users on the Plus plan.
I am a curious person; often I am an early adopter, often against intense friction (I was lambasted for being early on Vite and Tailwind in my tech career, and was likewise ridiculed for arriving too early with my flavor of music on Nashville’s Broadway - each of those are now mainstays).
Contrary to the popular derision of AI skeptics from AI hype boosters (the phrase “you just have your head in the sand” comes to mind), I was in fact an early adopter of ChatGPT. I was there from the beginning, watching it evolve into what it is now. I tried Cursor. I was using Claude and Claude Code while the general populace was still finding out about ChatGPT, and later switched my AI assisted workflows to Codex when a trusted colleague informed me of its edge in our domain. I have enjoyed (to an extent) using these tools as coding assistants, and I do so quite effectively. I even read some of the research papers.
While I have not bothered to use every available tool (keeping track of all of them could be a full time job unto itself), I have never denied the potential value of some of these tools as a reasonably competent programmer capable of reigning in their shortcomings (though I have strongly contended that the marketing narrative must align with reality). That said, I use Codex consistently enough to notice my usage limits significantly shrinking just within the past week, despite rather efficient utilization. After modest personal use over the weekend and half a day at my office, I was informed that I had less than 25% of my weekly Plus plan limit remaining. I have not changed my usage habits, save for implementing even more efficient techniques.
This comes as no surprise, of course. The token cost surge was inevitable, it was only a matter of when it would come.
AI winter might be coming sooner than predicted.
Good thing this news didn’t break before their recent $100 billion funding round.
Read part 1: “AI Boosters Are Rude (On AI Hype)”
Read part 2: “Differentiating AI Boosters From Optimists (On AI Hype pt. 2)”