AI Usage Limits Feel Like a Black Box

We already know that AI is costly, far more expensive than what most individuals used to pay for software subscriptions or even one-time software purchases. To some extent, that's understandable because we're constantly told that LLM providers are losing money. I'm not even talking about API costs, which are becoming almost unaffordable for individuals, especially when using U.S-based providers. All of that is acceptable, or at least something we can live with. What I don't understand is the complete lack of transparency around token usage. When you open Claude's usage page, the main thing you see is a push to upgrade to a higher-tier plan. There are no meaningful graphs, no detailed history, no breakdown showing what consumed your tokens, and no easy way to optimize your usage. And don't get me started on the hourly limitations, they're incredibly frustrating. Cursor isn't much better. You hit a limit, get a pop-up saying you've reached it, then open the usage page and find only a very high-level monthly summary. Could these companies make token usage more transparent and understandable for users? Of course they could. But it seems they don't want to. Part of the business model appears to rely on keeping pricing and consumption vague and difficult to track. You'll never know exactly where your tokens went. Was the high usage caused by your prompts, their internal orchestration, hidden context, agentic workflows, or something else? There's no visibility into it. The result is that we're expected to trust and pay for a service that we can't properly audit, measure, or even fully understand. That's a difficult position to be in when AI is becoming an essential tool for so many people.