4 minutes is how long my claude code agent took to ship the wrong thing the first time i caught myself skipping a spec the way i'd skip it with a senior eng i'd worked with for years.
"you get the intent." the agent didn't get the intent. with a person it would have been a slack message at 4pm asking the clarifying question. with the agent it was an artifact, fast, silent, wrong.
the failure wasn't the model. it was me importing a leadership instinct that worked for managing humans where the relationship fills the gap when the brief is fuzzy. there's no relationship to lean on with an agent. the spec has to do the work the relationship was doing.
solo builders run into the same thing harder because the team is just you. the instinct is "i know what i want, i don't need to write it down." with an agent that means the spec gets skipped because there's no second human in the loop to surface ambiguity.
what's the moment you first noticed your team-of-one instinct breaking with an agent?
The "you get the intent" instinct is exactly what breaks first.
I noticed it when I stopped writing out what I didn't want. With a person, half the spec lives in shared context — past decisions, things you've already ruled out, the general direction you're heading. With an agent, none of that transfers. Every session starts fresh, so anything you don't write down doesn't exist.
The shift that helped me most: treating the prompt less like an instruction and more like a handover note to someone who has never worked with you before. Because that's exactly what it is, every single time.
yeah - same thing hit me. I kept assuming what was ruled out was obvious. with a person it is. with an agent you have to write down what you didn't want, which means first knowing what it was.