You Are the Bottleneck
You use AI for drafts, summaries, and rewrites. It handles the easy 80 percent. But the hard 20 percent, the reviews, the sign-offs, the "is this good enough" calls, still has your name on it. The important work cannot move until you look at it.
The fix is not better AI. It is teaching AI to do the hard 20 percent the way you do it, against your standards, so the work stops piling up behind you.
The Four-Piece Prompt
Every judgment system is a prompt with four pieces. Most people write generic background and wonder why the output is generic.
The Copy-Paste Pack
The prompt that runs your judgment, the worksheet that pulls your standards out of your head, and the verdict format it should return.
The Judgment-System Prompt
[the standard you check first]
2. [the mistake you see most often]
3. [what "done right" looks like to you]
4. [add every standard you actually enforce. This list IS your judgment.]
Here is the work to review:
[paste the code, doc, proposal, or submission here]
TASK: Review this against my standards the way I would. Find what I would find. Do not be polite. Do not be generic. If it is genuinely clean, say so in one line.
FORMAT:
- VERDICT (one line): ship it / fix these first / redo it
- PROBLEMS, worst first: for each, what is wrong, which standard it breaks, and the smallest fix
- ONE THING a junior reviewer would miss that I would not
Tone: direct, no filler, no praise for effort. You are me on a busy day.Write Your Standards
Open a blank note. Pick one task where work piles up behind you. Answer these four prompts honestly. The answers go straight into the Context block above.
- The standard you check first. When this work lands on your desk, what do you look at before anything else?
- The mistake you see most often. What do people get wrong that you always have to fix?
- What "done right" looks like to you. Describe the version you would sign off on without changes.
- Every standard you actually enforce. Keep going. Specifics beat vagueness. "Numbers need a baseline to mean anything" beats "make sure it is good."
The Verdict Format
Force a one-line verdict first, so you can act in seconds. Only then the problems, worst first, each tied to the standard it breaks, with the smallest fix.
Keep It Sharp
One skeleton serves every review, and a 15-minute cadence stops it from drifting.
One Skeleton, Many Reviews
Save the prompt once. The same skeleton reviews different work on different days, just with different standards in the Context block.
Different standards, same shape. One system becomes many. The first build is the slow one. Every review after is fast.
Iterate Every Three Weeks
Your standards shift. The work changes. The output drifts, and you do not notice until it misses something obvious. Fifteen minutes every three weeks keeps it sharp. Three moves:
- Review. Run your system on fresh work. Is it still catching what you catch?
- Improve. Add the standard you learned since last time. Remove one that no longer matters.
- Expand. Build a second judgment system for the next bottleneck that piles up behind you.
Skip this and what you built dies in a month. Do it and it compounds. You stop being the bottleneck in more and more places.