Are flat-rate AI subscriptions structurally broken?

Anthropic is rolling out weekly rate limits for Claude Code on Pro and Max plans starting August 28. They're saying it'll only affect ~5% of subscribers, the ones running Claude Code continuously in the background, plus a handful violating ToS by sharing or reselling accounts.

Link to article here.

What's interesting to me is this is the third major AI coding tool in ~2 months to hit the same wall. Cursor and Replit both made similar pricing changes in June. Feels like flat-rate $20/month subscriptions are structurally incompatible with agentic workloads that can burn compute 24/7. The unit economics just don't math.

A few things I'm chewing on:

• Is usage-based or credit-based pricing now inevitable for any AI product with an agent layer?

• Anthropic claims Max 20x gives 20x the usage, but the disclosed numbers work out closer to 6x on Claude Code hours. Are we entering an era where "20x" means tokens, not outcomes?

• Cursor got dragged for communicating their pricing change poorly. Anthropic seems to be getting ahead of it with the "<5% of users" framing. Does that land, or does it read as spin?

Curious what others building or pricing AI products are taking away from this.