The Human Is the Loop: A Builder's Case Against Reflexive AI Dependence

After a few weeks offline on vacation, a developer returns to eleven terminal tabs of paused agents and a stack of unread Claude chats spanning taxes, landscaping, and policy review — and reads them not as productivity but as accumulated guilt. His argument is that heavy agent use had quietly made him less curious, less confident, and a little depressed. The tools became a buffer he slid between himself and stressful tasks, and a driver of pointless work: half-finished side projects that run but help no one, dictated ‘brainstorming’ sessions with a sycophantic chatbot he never trusted enough to actually use, and automation built mainly to justify building more automation — a ‘productivity ouroboros.’

He’s careful not to blame the technology itself. His read is that agents aren’t neurochemically addictive the way infinite feeds are, but they do produce real habituation: once AI is in one part of your workflow, more uses keep suggesting themselves. He also flags the incentive layer — a large share of the industry is betting on mass economic dependence on LLMs, since that’s the only story that justifies the valuations.

The piece isn’t anti-AI; it closes with a disciplined counter-example. Kicking off a real project, he scoped a single agent narrowly, loaded it with context, stated his own uncertainties, and asked for flagged gaps rather than a finished answer — accepting it won’t ‘10x’ him but might surface blind spots across messy systems he dislikes navigating. The takeaway, and the title: don’t let the human get trapped inside the agent loop. The human is the loop, and you tag the agent in occasionally and deliberately.