Use a one-in-one-out rule: every time you hand a recurring task to AI, retire the manual version and write it down as a briefed, assigned, reviewable job. If you haven't captured it that way, you haven't delegated it. You're just asking a chatbot the same question again next week.
This is the shift happening right now, and most people are missing it. The novelty phase is over. The folks getting real value out of AI aren't the ones with the cleverest prompts. They're the ones treating it like a team member they actually onboarded.
What's changing: from prompting to delegating
A year ago, the winning move was learning to write a good prompt. Now the tools are good enough that a decent prompt gets you a decent answer nearly every time. The bottleneck moved.
The new bottleneck is repetition. You write a solid prompt for your weekly update, get a good result, close the tab. Next Monday you're rebuilding the whole thing from scratch. You never delegated the task. You just did it faster, once.
Delegation means the task leaves your plate and stays gone. That takes a system, not a good moment.
Why the one-in-one-out rule fixes this
Borrow the idea from decluttering: for every new thing that comes in, one old thing goes out. Applied to AI, it means every task you hand off gets logged as a real, repeatable job, and the manual version gets retired from your head.
The win isn't the AI. It's that you stop holding the task in working memory. When "write the client recap" becomes a briefed job you can reuse, your brain lets it go. That's the real gain. Not speed. Release.
Without the rule, you accumulate. You keep re-explaining context, re-inventing the format, re-checking output you have no standard for. It feels busy and gives you almost nothing back.
The three parts every real handoff needs
For a task to leave your plate, the handoff has to survive being done again next month by a slightly different version of the AI. That takes three things.
1. A brief that travels
A good brief carries what the AI can't guess: who the output is for, the tone, the constraints, what "done" looks like. The trap is keeping this in your head and re-typing a rough version each time. Write it once. Store it somewhere you can grab it. A brief you can't find is a brief you'll rewrite.
2. An assignment you can point to
Naming the job matters more than it sounds. "Monday client recap" is a thing you can hand off. "Uh, that email summary thing" is not. When a task has a name and a home, you can hand it to AI, a teammate, or future-you without a five-minute setup every time.
3. A review step with a standard
This is the part everyone skips, and it's why delegation quietly fails. If you don't know what good looks like before you read the output, you'll either rubber-stamp junk or rewrite the whole thing. Decide your check in advance: two or three things that have to be true before you ship it. Review takes 60 seconds when you know what you're looking for.
How to run one-in-one-out this week
- Pick one task you've asked AI to do more than twice. Just one.
- Write its brief down properly, once. Not in a chat window. Somewhere permanent.
- Give it a name and a home so you can find it Monday.
- Write your review standard: what makes this output shippable.
- Then retire the manual version. Stop keeping it in your head.
Do one task a week. In two months you'll have a small stack of jobs that run themselves, and a noticeably quieter head. That's the part that compounds.
Where a system beats willpower
You can build this yourself in a notebook or a Notion page. The honest problem is that most people write the brief, skip the review standard, and lose the whole thing within a week. The structure is what makes the habit stick.
If you'd rather start from something ready-made, I built The AI Delegation Pack: Brief, Assign, and Review to hold exactly these three steps in one Notion space, so each task you hand off gets captured and stays captured. It's the shortcut if building the system yourself is the thing that keeps not happening.
Either way, the principle holds: don't just prompt faster. Hand the task off for good.
FAQ
What does the one-in-one-out rule mean for AI tasks?
For every recurring task you hand to AI, you retire the manual version. You write it up as a briefed, named, reviewable job and stop holding it in your head. One task delegated, one thing off your mental plate.
Isn't a good prompt enough to delegate a task?
A good prompt gets you one good result. Delegation means the task doesn't land back on you next week. That takes storing the brief, naming the job, and setting a review standard, so it's repeatable and not just fast once.
How many tasks should I hand off at once?
One at a time. Pick a task you've already asked AI to do more than twice, set it up fully, then move on. Trying to systemize everything in a weekend is how the whole thing collapses.
Why does the review step matter so much?
Without a standard for "done," you either approve weak output or rewrite everything yourself, which means you never really delegated. Deciding two or three checks in advance keeps review fast and quality honest.