The middle ground
The first lesson named the clear yes-delegate cases and the clear no-delegate cases. Most real tasks sit in between. This lesson is a rubric for that middle.
There are three dimensions to weigh:
Dimension 1 — Stakes
How bad is it if the output is wrong?
- Low: a typo in my draft.
- Medium: a flawed analysis in a blog post.
- High: a factual claim in a grant application to a funder.
- Very high: advice to a student about what to major in; copy in a public health campaign; the wording of a layoff announcement.
Higher stakes don't mean "never delegate." They mean "delegate less, review more, document the reasoning."
Dimension 2 — Reversibility
If the output turns out to be wrong, how much of a mess is it to clean up?
- Fully reversible: delete the draft and try again.
- Partially reversible: a public tweet you can delete, but screenshots exist.
- Irreversible: a decision someone acted on; an email already sent; a commit pushed to a shared branch.
The reversibility axis is underrated. A medium-stakes but irreversible task should be treated like a high-stakes one.
Dimension 3 — Your ability to judge the output
The cruelest part of AI delegation: the AI is best at tasks you are least equipped to check. If you're an intermediate writer, you can spot a bad paragraph. If you're a beginner in a technical field, you probably can't.
- You can judge it easily — delegate freely, review briefly.
- You can judge it with effort — delegate, schedule real time to review.
- You can't reliably judge it — don't delegate alone. Either learn enough to judge, or have someone else review.
The rubric
Put the three together and you get a small decision table:
| Stakes | Reversibility | Your ability to judge | → Move | |---|---|---|---| | Low | Any | Any | Delegate fully | | Medium | Reversible | Easy | Delegate, skim | | Medium | Reversible | Hard | Delegate, real review | | Medium | Irreversible | Any | Delegate, review, slow down | | High | Any | Easy | Delegate, treat as "assistive" | | High | Any | Hard | Don't delegate alone | | Very high | Any | Any | Keep; use AI to critique your own draft |
This isn't a hard rule. It's a starting point. The useful move is noticing which row you're in before you delegate.
Live practice
Take the 3 "wavering" tasks from last lesson's homework and place them on this rubric. Notice which row each falls into. Write down what you'd do differently in each case — how you'd delegate, what review you'd do, who else would be in the loop.
Share one of them on the discussion thread under this lesson. The point isn't to be right; it's to practice the noticing.
Inspired by Anthropic's "AI Fluency: Framework & Foundations". The three-dimension rubric is a common synthesis; the particular table here is original.