Delegating to AI: when to do a task yourself and when to hand it off

Delegate a task to an AI agent when it's clear, bounded, and reversible — and keep it yourself when it needs your judgment, carries real stakes, or is hard to un-do. That single test covers most decisions. The mistake people make isn't delegating too much or too little; it's delegating without a rule, then either micromanaging or getting surprised.

Hand it to an agent when…

• You can write a two-line brief a competent stranger could follow.
• The output is easy to check — a draft, a list, a summary, a first-pass.
• A wrong result is cheap to fix or throw away.
• It's repetitive or high-volume and drains time better spent elsewhere.

Keep it yourself when…

• It needs taste, relationships, or a judgment call only you can make.
• The stakes are high and a mistake is expensive or hard to reverse.
• You can't yet describe what "done" looks like — figure that out first.

The in-between: delegate with a gate

Plenty of work is worth an agent's first pass but too important to close unseen. That's exactly what an approval gate is for: let the agent do the work and propose a result, but require your sign-off before it counts. You get the speed of delegation and the safety of review — the best of both.

Get better at briefing

The real skill isn't picking tasks — it's writing the brief. Say what done looks like, what to avoid, and what context to use. The clearer the brief, the more you can safely hand off. Over time, the line moves: work you once kept becomes work you delegate with a two-line note.