I've been a Max user of Claude all year and it's crucial to my day to day work. Last night I had a new experience which makes me feel like a junior dev who blew thousands of dollars on a misconfigured AWS CDK file.To avoid agents doing online research from leaking private info I created a persona which can do online research as a subagent from a fresh context, I asked it to fact check claims as well. Little did I know subagents don't report their token usage and my subagents ended up recursively checking claims. I burned through 80% of my Fable usage for the week in 10 minutes.I've never come close to using my full token allowance before and now I'm going to have to spend money on expensive usage tokens or upgrade to 20x (or change to a different provider) to complete this weeks work.Anthropic's online support agent was terse, "empathizing" but ultimately said it could not do anything.Wouldn't the right thing from a Customer Support perspective to give a one-time usage reset in a case like this? If they had a human instead of an agent doing it, would that have changed the outcome?I've implemented bounding now in my persona's but it just seems like this is a preventable problem that they're (1) not solving and (2) not willing to remedy when it did happen.Full disclosure, I love what Anthropic has built in Claude and use the product daily. Comments URL: https://news.ycombinator.com/item?id=50252272 Points: 1

Comments: 0