AI agents became his procrastination ritual, not his productivity engine

A few AI-free weeks revealed the crutch: agents were doing the feeling of progress, not the work

A personal essay in which the author diagnoses his own AI habit as avoidance dressed up as productivity — and proposes a narrower, intentional use instead

Judgmentauthor's own retrospective judgment, single case

The author's core claim is not that AI agents are bad, but that his own usage pattern — many parallel agent threads, dictation-to-chatbot ‘thinking partners,’ automation for its own sake — was a habit that made him less curious, less confident, and probably a little depressed, while producing nothing anyone needed.

What broke the pattern

A multi-week vacation without laptop or phone removed AI entirely. Returning to eleven paused cmux tabs and unread badges across taxes, landscaping, and policy review, he didn't feel behind — he felt relief, and then recognized the backlog as ‘stubs of guilt,’ not progress.

The habit he's diagnosing

What he told himself

  • Bouncing ideas with ChatGPT while driving was productive use of otherwise-dead time
  • Setting up automation would make him a faster, more parallel builder
  • More agent threads meant more gets done

What he now thinks was actually happening

  • It was a sycophantic mirror he never trusted enough to use the output of
  • No one asked for the efficiency gains; the tool became the task (a ‘productivity ouroboros’)
  • Each open thread was really an agent inserted as a buffer between him and a task that caused stress
  • The learning — the actual source of enjoyment in past tinkering — got skipped to reach a result

The one use he still endorses

kept

Before writing this post, he gave an agent (pi) a narrow, well-specified task: scan codebases, wikis, schemas, and conversations for a work project, flag issues, propose options — with his own context and open questions included.

why it's different

It's constrained, single-purpose, and pattern-matches across systems he doesn't enjoy navigating himself — rather than replacing the thinking or writing he wants to do directly.

explicit non-claim

He states plainly: it won't hand him the perfect solution and won't ‘10x’ him — its value is narrower, freeing time for the writing and reflection he's doing right now.

The industry backdrop he flags but doesn't develop

He notes, almost in passing, that a large part of the tech industry's valuations depend on users becoming ‘hopelessly dependent’ on AI tools — implying his personal habit may be less an accident than a designed outcome of product and business incentives. He doesn't substantiate this with data; it's asserted as a hunch alongside his belief that agent use isn't neurochemically addictive like feed-scrolling.

Stated plan going forwardstated intention, untested

Be intentional about when to use AI and when not to, and be honest about what is actually gained versus lost each time — narrow, well-scoped tasks (like the pi example) rather than open-ended parallel delegation.

Sources