Decision Abdication: The Hidden Risk of Everyday AI Use

Nobody talks about the part where AI quietly takes over your decisions. It doesn't happen dramatically. It happens through small moments of relief and small moments of fatigue, until one day you realise you've stopped being the author of your own work — you're just approving the machine's.

This is decision abdication, and it shows up in two opposite-looking forms.

Two ways we abdicate

Abdication through relief. You're busy. The first AI output looks fine. You accept it without really evaluating it, because evaluating it takes energy you don't have right now. The decision wasn't made — it was skipped.

Abdication through perfectionism. The opposite trap. You iterate endlessly, asking for revision after revision, but without any clear criteria for what "done" looks like. You feel busy and rigorous, but you've still handed the standard to the model, because you never set one yourself.

The habit that keeps you the author

The fix is the same in both cases: define your criteria before you prompt, then make a deliberate human edit after. Three scenarios show how it works:

Writing a job description. Before prompting, decide the 3–5 core responsibilities, the real team culture, and the appropriate experience level. Generate, then review against those standards and make one focused edit.

Planning a client presentation. Set the requirements first — problem-first framing, no more than three main points, a clear next step. Then check that the data exists and the logic actually sequences.

Drafting a policy document. Pre-define the standard — readable by non-experts, addresses real scenarios, stays concise — and protect the organisation-specific details AI can't know.

Three rules to live by

Limit yourself to three prompts per task, so you can't spiral. Always make an intentional human edit after the AI produces something usable. And set your criteria before prompting, never after. These small disciplines build what matters most: epistemic habits — patterns of thinking that let you leverage AI without surrendering to it. Success isn't when the AI is happy. It's when the output meets your criteria and you've protected what matters.