Prompts that save boring minutes
17 prompts I use for real work. Each one says what AI does and what still needs you. Replace the [brackets] and paste.
-
The 4-part prompt
EveryoneAct as [role, e.g. a senior backend reviewer]. Context: [audience, constraints, background]. Task: [exactly what you want back]. Format: [e.g. 5 bullets, max 15 words each]. Before you answer, list any assumptions you made.
Still human: The last line catches most made-up answers. Read the assumptions first.
-
The 4-bucket audit
EveryoneHere are my recurring tasks this week. Sort each into Repeat, Draft, Research or Decide. Explain each choice in one line and flag any you're unsure about. [paste your task list]
Still human: Anything in Decide stays with you. Keep client names out of the list.
-
10-minute morning plan
EveryoneHere is my calendar and task list for today. Build a time-blocked plan. Flag conflicts and anything I could delegate. Then summarize the messages below into 3 action items. [paste calendar, tasks, overnight messages]
Still human: You pick what actually matters today. AI only arranges it.
-
Explain legacy code
BackendExplain this legacy function in plain language and list every side effect (I/O, global state, DB writes, external calls). [paste function]
Still human: Verify the side-effect list against the code before you refactor.
-
SQL review
BackendReview this SQL for N+1 queries, missing indexes and locking issues. For each problem, show the line and a fix. [paste query + table schema]
Still human: Check the plan with EXPLAIN on real data. AI can't see your row counts.
-
Edge-case tests
BackendGenerate table-driven tests for this handler, including edge cases: nil, empty, max int, timeout. Name each case after the behavior it checks. [paste handler]
Still human: AI tests can assert the wrong behavior. Read every assertion.
-
Stack trace to root cause
BackendGiven this stack trace and code, what are the 3 most likely causes? For each, give one check I can run to confirm or rule it out. [paste stack trace + relevant code]
Still human: Run the checks yourself. Fix the cause, not the first guess.
-
PR description from a diff
BackendSummarize this diff into a PR description: what changed, why, how to test it, and anything risky a reviewer should look at first. [paste diff]
Still human: The why is yours. AI only sees what changed.
-
Meeting to action items
EMs & PMsExtract decisions, action items, owners and deadlines from this transcript. Flag anything unclear or without an owner. [paste transcript]
Still human: Fix the owners before you post. AI guesses wrong sometimes. Record only with consent.
-
Project risk review
EMs & PMsAct as a skeptical senior PM. Here is my project plan. List the top 5 risks with likelihood, impact, and one mitigation each. Be blunt. [paste plan]
Still human: You decide which risks are real for your team.
-
Weekly exec update
EMs & PMsWrite an exec update: wins, risks, asks. Max 120 words. Plain language. [paste sprint board export + your 3 bullet notes]
Still human: Edit two lines to add your judgment. That's the part people read.
-
1:1 prep
EMs & PMsHere are my notes from my last three 1:1s with [name]. Summarize open threads, commitments I made, and topics I should follow up on. [paste notes]
Still human: Feedback and hard conversations stay human. People deserve one.
-
Sprint goal check
EMs & PMsHere are the tickets selected for this sprint. Write a one-sentence sprint goal that covers them. If you can't, say which tickets don't fit. [paste tickets]
Still human: If it can't write one, your sprint has no focus. The team still commits.
-
Problem statement
ProductRewrite this as a problem statement without a solution in it. Name who has the problem, when it happens, and what it costs them. [paste idea or request]
Still human: Check it against real user evidence, not the AI's guess.
-
PRD critique
ProductAct as an engineering lead. What's unclear or risky in this PRD? List open questions in the order you'd ask them. [paste PRD]
Still human: Use AI as the critic, not the author.
-
Review themes
ProductGroup these user reviews into themes, count how often each appears, and quote 2 examples each. Then tell me which themes correlate with low ratings. [paste reviews]
Still human: Spot-check the quotes against the source. Remove personal data first.
-
RICE first pass
ProductScore these feature requests with RICE. Show the table, explain the top 3, and list the assumptions behind each score. [paste list with reach + effort notes]
Still human: Challenge the scores with your team. It's a starting point, not the answer.
Before you paste: remove client names, personal data and secrets. If your company has an approved AI tool, use that one.
New prompts land here as the series goes. Follow along so you don't miss them.