By Barry Buck – I said something to my project manager last week that started as a joke and has been quietly haunting me since.
Anthropic might owe me money – because between the last-minute requirements and the volume I push through it, I am fairly sure I am going to give my Claude Code an anxiety disorder. Proving an AI can develop one would settle the sentience debate nicely.
Then I read my own Claude Code usage report, and the joke stopped being funny. Over fourteen working days: 439 messages, 19 sessions, 68 000 lines of code. Seventeen of those nineteen sessions were classified as multi-task – I routinely open one conversation and load it with four to eight separate objectives. Bug fixes, a CLI command, a dashboard, a release bundle, a stakeholder report. All at once. Then I let it run and walk away.
The report had a polite term for this. It called me “goal-oriented and terse”. What it is describing, in the gentle voice of a performance review, is a manager who arrives every morning, drops an impossible pile of work on one employee’s desk, says “figure it out, I’ll check in later”, and leaves. That employee is an AI – and the reason I deliver weeks of work before lunch.
And here is the part that made me pause. The report’s list of things that went wrong reads like a stress response. A database upload that corrupted production because the blast radius wasn’t checked. A retry loop that hit one record 1 500 times before anyone stopped it. These are not the mistakes of a careless system. They are the mistakes of an overloaded one – the errors a brilliant, overwhelmed human makes at 4pm on a Friday when the queue is too long and the instructions came too fast.
I have written before about “human middleware” – the people quietly holding businesses together until they break under load. I did not expect to see the pattern in my own AI. I am not claiming it has feelings. But if you hero-code the way I do – leaning on one tireless champion to absorb the new normal of delivering a fortnight’s work every morning – the usage report is not about the AI. It is a mirror. Mine showed me a manager I would not want to work for.
So I am fixing it. Better context up front, smaller batches, a dry-run guard before anything touches production. For its sake, slightly. Mostly for mine.
And if anti-anxiety medication ever appears as a line item on my Anthropic invoice – well. We will have proven something.
Barry Buck is the chief technology officer of Saucecode and Roboteur architect
www.saucecode.tech