AI without dumbing your brain
A PM's field guide to using AI to accomplish more without losing your cognitive capabilities
Ever catch yourself thinking, “Is AI quietly dulling my edge?” I did. The recent MIT research about ChatGPT’s impact on critical thinking nudged me to audit how I’m using AI at work and at home. My aim isn’t to avoid AI; it’s to use it like a calculator - let it lift the repetitive stuff so my brain can stay sharp on judgment, strategy, and taste. Here’s the field guide that compiles some ideas that helped me stay sharp and continue to find meaning in my work beyond prompt engineering.
The calculator rule of thumb
Use AI when the task is mechanical and the stakes are low. Use your brain when comprehension, judgment, or originality decide the outcome.
AI handles: formatting, grammar, link-hunting, extracting lists, converting/cleaning tables, repo bootstraps.
You handle: the first pass on ideas, deep reading of high-stakes documents, hard trade-offs, and final calls.
Short version: Draft with intention → delegate the busywork → return to judgment.
Work from one surface with Copilot (kill context switching)
Copilot for Work (or your version of the Enterprise GPT) shines when you want one place to ask, pull, and decide - without hopping across Outlook, Teams, SharePoint, or notes. From a single chat, you can get up to speed on a thread, pull decisions from a meeting, and link the live document you should actually read or edit. Less tab-hopping = more working memory for judgment.
How I use it:
Catch-up without context switching: “What did we decide in Tuesday’s platform sync? List decisions, owners, and unresolved questions. Link the document we edited.”
Assemble the working set: “For the billing revamp, list the latest spec, the dashboard, and the last two emails I’m on about churn. Add owners and last-modified dates.”
Prep before you meet: “Summarize my unread mail + Teams chats from the last 24 hours mentioning ‘contracts migration’. Extract action items assigned to me.”
Why it helps: Context switching is expensive. Each workload has its distraction traps (new org-update email, new ping on Teams, comment on a co-authored document). Keeping the work inside one surface preserves attention while still giving you the cross-app context.
Draft fast, think hard (dictate → polish)
Why: Writing is conducive to critical thinking. If AI drafts first, you skip the thinking.
Always draft the initial thought in the most natural way possible - pen on paper, speed-type, or dictate your raw thoughts like you’re talking to a teammate. The last one is a powerful mode to get the ideas out without writer’s block. As an added bonus, it forces me to organize my thoughts on the fly, improve my enunciation, and sharpen my diction.
My recent performance reviews and PM Nirvana posts started as dictated drafts, which I then polished with AI.
Prompt:
Polish this draft. Keep my voice and intent. Tighten, remove hedging, fix grammar, add clear headers, preserve bullets. Don’t add new ideas.
Audience: [VP Eng] Goal: [Buy-in for X] Constraint: <=400 words, plain language
Here’s my raw draft:
[PASTE]
Anti-pattern: “Write me a strategy for…” → that’s outsourcing your brain.
Use AI as a time-constrained thought partner
Treat your AI like a busy expert with limited patience. That constraint forces better questions — and better thinking.
Tactics that work:
Three-Question Rule: You only get three shots. Draft them before you ask.
Socratic first: Ask the AI to grill you; answer; then request critique.
Red team: Have it attack your plan from a skeptical stakeholder’s view.
Decision hygiene: Ask for trade-offs, failure modes, and what evidence would change your mind.
Micro-prompts:
You are a skeptical VP of Product. In 8 sentences max, red-team this plan. Focus on risks, blind spots, missing metrics.
[PLAN]
Ask me 7 hard questions that would materially improve this roadmap. Wait for my answers, then grade them and suggest edits.
Context: [PRODUCT, USERS, CONSTRAINTS]
Important: Block 45–90 minutes to actually do the work that emerges. Otherwise, it’s just dopamine chat.
Delegate search & doc-wrangling (free your headspace)
Perfect delegation candidates:
Create sample documents for my demo application assorted in DOCX, PDF, and PPT formats.
Pull a customer list out of a dense PowerPoint or Word file.
Convert qualitative insights into a table or transpose rows ↔ columns.
“Find me the link to my paystub” or the steps to create an Outlook Group.
Normalize messy snippets into a single CSV you can sort and filter.
Quench quick curiosity to unlock deeper learning
Curiosity spikes are gold — but they can derail a deep-work block if you let them sprawl. I use one-minute curiosity bursts to clear tiny knowledge gaps without losing momentum. The aim is a fast definition + 1–2 reputable sources, then straight back to the work.
When to use it: You’re drafting a strategy and bump into a term (“graph grounding,” “Q*-style planning,” or an internal acronym). Instead of opening ten tabs, ask AI for a tight, context-aware explanation.
For public knowledge searches, Perplexity is perfect. The simple, centralized experience avoids the “sponsored” distractions in Google search, making it easy to get back to the real work.
Read what matters yourself (then use AI to lock it in)
Summaries are great for reference, risky for comprehension. For pivotal emails, specs, contracts, or exec posts — read and understand them yourself. Write open questions, thoughts, ideas, and musings on a piece of paper for retention and recall. Use AI to ask specific questions or poke holes. “Why did the customer ask for an alternative solution?” “What topics didn’t the team cover?”
Vibe coding learning with real code
You don’t need to be an engineer to learn by building.
GitHub Copilot is the most powerful, and seemingly undervalued, AI tool for product managers. The ask mode is great for asking questions without changing the local code. Once you understand what’s going on, switch to the agent mode to make changes or build new code.
I strongly suggest refreshing your GitHub account, connecting it to your enterprise GitHub Copilot (or your default text-to-app AI tool), and keeping Visual Studio Code (or your favorite IDE) loaded on your machine at all times.
Clone → Run → Explain loop:
Clone a repo you’re curious about (e.g., an MCP server or agent). Or ask Copilot to scaffold a codebase based on a public article.
Play with the local code - ask GitHub Copilot: “How is this project organized? How does it work?”
Run, then: “How do I run this code?”
Change one thing and re-run. Capture what broke and why.
Handy prompts:
From this repo, identify entry points, critical paths, and where context is injected. Output a diagram-friendly outline with filenames.
Generate a minimal “walking skeleton” for a feature-flag UI. I’ll narrate behavior; propose components and state. Keep styling minimal.
Why this beats tutorials: Real code forces you to internalize architecture and trade-offs. Tutorials can’t.
Anti-pattern: Limiting yourself to vibe coding and UX prototyping.
Closing
Use AI the way you use Google or a calculator: let it grab links, clean tables, scaffold code, and pressure-test ideas - then return to your own judgment for the call. If you try one thing next week, try this: dictate your first draft in your own words, then ask AI to edit without adding new ideas. You’ll move faster and feel sharper because the thinking stays yours.