UNPKG

major-ai-skills

Version:

Installable agentic skills / AI agent skills (SKILL.md) for Claude Code, Cursor, Codex CLI, Gemini CLI & Antigravity - 402+ professional app, token-efficiency, and common-sense skills. SEO/GEO ready.

86 lines (62 loc) 4.22 kB
--- name: think-of-ai-as-a-junior-assistant description: "Delegate a bounded task with context and acceptance criteria, then review the result before relying on it." category: common-sense risk: safe source: self source_type: self date_added: "2026-08-26" tags: ["mental-models", "delegation", "quality-control", "management", "productivity", "prompt-engineering"] tools: ["claude", "cursor", "gemini", "codex", "chatgpt"] --- # Think of AI as an Eager Junior Intern (The Delegation Model) (AI Skill) ## Overview Users typically fall into two flawed extremes with AI: 1. **The Oracle Fallacy**: Treating the AI as an infallible genius and blindly trusting every hallucinated number and URL without review. 2. **The Cynic Fallacy**: Dismissing AI as useless because it made one factual error. The **Eager Junior Intern Mental Model** frames the relationship accurately: imagine you have an exceptionally well-read, lightning-fast intern who has memorized all of Wikipedia and GitHub, but **lacks real-world judgment, common sense, and accountability**. --- ## The Junior Intern Delegation Matrix ``` ┌─────────────────────────────────────────────────────────────┐ Junior Intern Management Loop [ WHAT YOU WOULD TELL A JUNIOR INTERN ]: 1. Don't guess; ask if you don't know 2. Use this exact template and follow these examples 3. Show me your outline before you write 20 pages 4. Double-check all math on a calculator before showing me Apply the EXACT same management rules to your AI prompts! └─────────────────────────────────────────────────────────────┘ ``` --- ## The 4 Rules of AI Delegation ### 1. Give an Explicit Rubric (Don't Assume Unstated Context) A human manager would never tell an intern *"Go do marketing"*; they say: *"Extract the pricing of our top 3 competitors into a spreadsheet by 3 PM"*. Apply the same specificity to prompts. ### 2. Set Up Checkpoints (Progressive Review) Review outlines and wireframes before letting the intern draft 1,000 words. ### 3. Require Proof of Verification (Show Your Work) Instruct the model to cite exact source clauses or write Python calculation blocks. ### 4. Retain Senior Sign-Off (The Final QA Gate) You are the senior manager. You own the final deliverable. You never send work to clients without reading and signing off. --- ## Master Delegation Prompt Template ```markdown I am assigning you a task as my Junior Research Associate: [DESCRIBE TASK]. Guidelines: 1. Follow this exact format schema: [PASTE TEMPLATE / HEADERS]. 2. For every claim, provide the source reference. 3. If you encounter missing data or ambiguity, do NOT guess. Mark as `[UNVERIFIED: Reason]`. 4. Keep the draft concise so I can review it in under 2 minutes. ``` --- ## Real-World Comparison ### Scenario: Competitor Pricing Analysis #### The "Oracle" Mistake (Blind Trust) > User asks AI for competitor pricing $\rightarrow$ AI invents a plausible but wrong $49/mo plan $\rightarrow$ User presents it to the CEO $\rightarrow$ Client corrects them in the meeting (Reputational damage). #### The "Junior Intern" Approach (Managed Verification) > User prompts: *"Find the pricing tiers for [COMPETITOR]. List each tier, its price, and the exact source URL/screenshot date. If pricing is hidden behind an enterprise sales form, explicitly state: 'CUSTOM QUOTE REQUIRED'."* $\rightarrow$ User clicks the 3 URLs in 30 seconds to verify $\rightarrow$ Delivers 100% accurate report to leadership. --- ## Summary Best Practice > **"Never ask an AI to do something you wouldn't trust a smart, first-week intern to do without your final review."**