# 25 Copy-Paste Prompts That Make AI Agents Genuinely Useful URL: https://whatisagenticai.net/blog/prompts-for-ai-agents/ Published: 2026-06-05 Publisher: Little Tough LLC, maker of AI Agent Assistant: Agentic AI (https://apps.apple.com/us/app/ai-agent-assistant-agentic-ai/id6761501260) Field-tested prompts for writing, coding, research, studying, planning, and multi-agent teams — with the one structural trick that makes agents deliver. Most prompt lists are written for chatbots: clever one-liners that produce one clever response. Agents deserve better briefs. The prompts below follow one structural rule, and it’s the only prompt-engineering rule that survives contact with 2026 models: **The rule:** give an agent a **goal**, the **constraints**, and what **“done” looks like**. Goal — context — deliverable. Everything else is decoration. Copy any of these straight into your agent of choice. They’re written for the expert agents in [Agentic AI](https://whatisagenticai.net/) but work anywhere. Tweak the bracketed parts. ## Writing (Writing Coach) ### 1. The email in your voice Draft an email to [person] about [situation]. Tone: warm but direct. It must (1) acknowledge their last message, (2) clearly say [the hard thing], (3) end with one concrete next step. Under 150 words. Then show a 50-word version. ### 2. The de-corporate-izer Rewrite this so a human wrote it. Kill buzzwords, passive voice, and throat-clearing. Keep all facts. Halve the length, then list what you removed and why: [paste text] ### 3. The cover letter that isn't generic Here's the job posting: [paste]. Here's my experience: [paste resume bullets]. Write a cover letter that maps my 3 strongest matches to their 3 biggest needs. No "I am writing to express my interest." Open with a claim I can defend in an interview. ### 4. The disagreement, defused I need to push back on [decision] with [person senior to me]. Draft a message that (1) steelmans their position first, (2) raises my concern with one concrete example, (3) proposes an experiment instead of a verdict. Professional, zero passive aggression. ### 5. The bio you've been putting off Interview me with 6 questions to extract what's interesting about my work, then write three bios from my answers: 1 sentence, 50 words, and 150 words. Confident but allergic to self-importance. ## Code (Code Helper) ### 6. The bug, actually explained Here's the error, the code, and what I expected: [paste all three]. Before proposing any fix: explain the actual cause in plain English. Then give the minimal fix, then (separately) the proper fix if they differ. Flag anything nearby that will break next. ### 7. The code review you'd pay for Review this function like a senior engineer who cares: correctness first, then edge cases, then readability. Rate severity (blocker / should-fix / nit). Don't rewrite it wholesale — show minimal diffs: [paste code] ### 8. The "explain this codebase fragment" Explain what this code does to a competent developer who's new to this codebase: first the one-paragraph summary, then line-by-line for the tricky parts only. End with: what would surprise a maintainer? [paste code] ### 9. The regex (or SQL) hostage negotiation Write a [regex/SQL query] that [exact requirement]. Then prove it: show it running against 5 test cases, including 2 designed to break it. If any fail, fix and re-run before answering. ### 10. The refactor with receipts Refactor this for readability without changing behavior. List every behavioral assumption you're relying on. If any assumption is risky, stop and ask me first: [paste code] ## Research (Research Analyst) ### 11. The meeting brief I'm meeting [person/company] tomorrow about [topic]. Build me a one-page brief: 5 things I must know, what changed in their world in the last 12 months, 3 smart questions to ask, and 1 landmine to avoid. Mark anything you're unsure about. ### 12. The honest comparison Compare [option A] vs [option B] for [my specific situation]. Build the criteria list first and ask me to confirm it. Then score both, show your reasoning per criterion, and tell me which single criterion should dominate my decision. ### 13. The steelman machine I believe [position]. Argue the opposite as compellingly as an informed, good-faith expert would. Then list the 3 pieces of evidence that would most change my mind, and where I'd find them. ### 14. The "what am I missing" Here's my plan: [paste]. Don't praise it. List the failure modes in order of (probability × damage), the assumption I'm most likely wrong about, and the cheapest test that would expose it within a week. ## Learning (Study Buddy) ### 15. The first-principles ramp Teach me [topic] from first principles. Rules: one concept at a time, check my understanding with a question before moving on, use analogies from [domain I know], and tell me when something is a simplification. ### 16. The exam drill sergeant My exam on [subject] is [date]. Quiz me with escalating difficulty. When I'm wrong: explain why from first principles, then re-test the same concept disguised differently 3 questions later. Track my weak spots and summarize them at the end. ### 17. The paper, digested Summarize this for three audiences: my boss (3 bullets, decisions only), me (one page, including methods and caveats), and a curious teenager (one paragraph). Flag the single most load-bearing claim: [paste text] ## Planning (Life Planner) ### 18. The Sunday reset Here's everything on my plate: [brain dump]. Build my week. Constraints: deep work before noon, gym ×3, [fixed commitments]. Rule: every task gets a day and a duration, or it gets explicitly cut. Show the cut list — don't hide it. ### 19. The goal, reverse-engineered I want to [goal] by [date]. Work backwards: monthly milestones, then this week's 3 actions, then today's first 25-minute step. Then tell me the most likely reason I'll quit in week 3, and design around it. ### 20. The trip that fits your actual family Plan [N] days in [city] for [group + ages]. Constraints: [budget], [diet], max 90 minutes per museum, downtime 2–4pm daily. Group by neighborhood to minimize transit. For each day: a plan A and a rainy plan B. ## Creative (Creative Brainstormer) ### 21. The name generator with taste Generate 20 names for [thing]. Then be your own harshest critic: kill 15 of them and tell me why, in one line each. For the surviving 5: check pronounceability, spelling-over-the-phone, and what each one accidentally implies. ### 22. The hook factory Here's my topic: [topic]. Write 10 opening lines for [format — post/video/talk]: 3 contrarian, 3 story-led, 2 data-led, 2 questions. No clickbait I can't cash. Mark which one you'd bet on and why. ## Multi-agent team sessions These are written for a [team of agents](https://whatisagenticai.net/blog/multi-agent-teams/) working one goal together — in Agentic AI, build a team of up to four and paste one brief: ### 23. The launch package Team goal: launch [product] on [date]. Researcher — who's the buyer and what nearly-identical thing failed before? Strategist — positioning and the one channel we'll win. Writer — landing page copy + launch email from that strategy. Planner — 3-week calendar, working backwards. Build on each other; don't repeat each other. ### 24. The decision panel Team goal: should I [big decision]? Analyst — lay out the evidence both ways. Skeptic — attack the optimistic case hard. Planner — assuming yes, what do the first 30 days look like? Then together: give me the 3 questions that actually decide this. ### 25. The content engine Team goal: one month of content about [topic] for [audience]. Brainstormer — 12 angles, no duplicates. Writer — turn the best 4 into drafts in my voice (sample attached). Planner — sequence all 12 with dates and formats. Flag the 2 ideas most likely to flop, and why. ## Why these work Look back at the structure: every prompt states a goal, gives real constraints, and defines the deliverable — and the best ones add a **self-check** (“prove it,” “show the cut list,” “kill 15 of them”). That self-check line is you borrowing the core mechanic of the [agent loop](https://whatisagenticai.net/blog/how-ai-agents-work/): act, then observe, then revise. Prompts that build the loop in get agent-quality results even from a single agent.