The AI power user's new constraint isn't creativity or skill. It's their token budget.
The Summary
- Anthropic tightened Claude's usage limits during peak hours, forcing heavy users to restructure when and how they work with AI
- About 7% of Claude users now hit session limits they previously didn't encounter, despite efficiency improvements meant to offset the impact
- Workers are adapting by breaking projects into smaller chunks, planning workdays around invisible usage meters, and treating AI access as a finite resource to be rationed
The Signal
This is the first crack in the "infinite intelligence" promise. For two years, the AI narrative has been: you can think bigger because the machine never gets tired. Now the machine is telling people it needs a break.
Max Johnson's workflow shift tells the real story. His startup Briix helps small businesses deploy AI, which means he's a power user, not a casual dabbler. He used to work in single, hours-long Claude threads, building context as he moved from social scripts to graphics to documents. That continuous dialogue burned tokens but produced sharper outputs. Now he's hitting limits two prompts into a fresh chat.
The economics are showing through the interface. Running these models costs real money, and companies like Anthropic are managing demand the same way utilities manage peak loads. They're not raising prices outright. They're rationing access during business hours when everyone wants to use the tool at once.
"You plan your day around knowing that you can spend X amount of time."
Here's what most coverage is missing: this isn't a bug, it's the business model asserting itself. Anthropic says only 7% of users are affected, which sounds small until you realize those are likely the 7% who generate the most value from the tool. The long-context power users. The ones building businesses on top of Claude's reasoning. The ones who were working "all hours of the day with no issues."
They're the canaries. If you're hitting limits now, you're either doing something very right or you're about to get priced into an enterprise tier you weren't budgeting for.
The Implication
If you're building a business on AI infrastructure you don't control, price the constraint into your model now. Usage limits will only get more sophisticated. Expect dynamic pricing, peak-hour throttling, and tiered access that mirrors AWS more than it mirrors ChatGPT's early days.
For individual users, this is a forcing function. The ones who adapt will get better at prompt compression, context management, and knowing when to use the expensive tool versus the fast one. The ones who don't will complain about AI getting worse when really it's just getting real about costs.