5 ways to use AI as a thinking partner, not a crutch

Key Takeaways

• Ask AI to challenge your ideas rather than validate them
• Use voice-to-AI tools for bionic dictation that organizes your thoughts
• Role-play with AI advisors to surface blind spots before major decisions

Most people use AI to do their thinking for them. Jeremy Caplan, who directs education at CUNY's Newmark Graduate School of Journalism, argues you should flip that model. In his Wonder Tools newsletter, he outlines five tactics for using AI to think harder, not less. The core idea: treat AI as an intellectual sparring partner that pokes holes in your reasoning, not a ghostwriter that lets you check out.

Why this matters for builders

Microsoft's 2024 Work Trend Index found 77% of employees now use AI at work. But a growing number of knowledge workers, around 65% in recent surveys, worry they're losing critical thinking skills to AI reliance. For product teams making architecture decisions or founders pitching investors, the quality of your thinking determines outcomes. Outsourcing that to a model is a liability.

Caplan's tactics aren't about avoiding AI. They're about using it to stress-test ideas before you commit to them.

Tactic 1: Ask AI to challenge you, not praise you

The default behavior of most AI assistants is to affirm whatever you say. That's useless for decision-making. Caplan suggests prompting explicitly for pushback. Present your plan, your reasoning, and your objective, then ask for devil's advocate responses.

His prompt template includes questions like: What are the strongest arguments against this approach? What risks might I be overlooking? Who else should I talk to? The goal is to surface blind spots before you finalize a presentation or ship a feature.

You can go further with role-play. Ask the AI to respond as a competitor, a skeptical board member, or a specific person you admire. Ethan Mollick, the Wharton professor and author of "Co-Intelligence," uses a council of advisors technique where he asks AI to simulate multiple perspectives at once. Microsoft's Copilot now offers a free "Team of Advisors" feature that animates this approach.

Tactic 2: Think out loud with bionic dictation

Some people think better by talking. Caplan calls this "bionic dictation": you speak a stream of consciousness, and AI transforms it into organized text.

Apple Notes and Google Keep handle basic mobile dictation for free. But tools like Letterly go further by summarizing your words and structuring them into outlines. This isn't about saving typing time. It's about externalizing messy thoughts so you can see patterns and gaps you'd miss in your head.

For product teams, this approach works well for post-standup brain dumps or capturing design thinking sessions. The output isn't the final artifact. It's raw material your brain then refines.

Tactic 3: Keep it private with offline AI

If you're thinking through sensitive decisions, sending them to cloud AI services may not be an option. Caplan recommends Jan, a free offline AI tool that runs locally. You get the sparring partner without the data exposure.

This matters for founders negotiating term sheets, engineers discussing pre-release architecture, or anyone whose thought process constitutes proprietary information. The thinking partner doesn't need to phone home.

Tactic 4: Use AI dialogue to surface what you forgot

Caplan shares a personal example: while discussing a new morning routine with ChatGPT, the back-and-forth revealed several things he'd left out of his planning. The AI didn't invent the missing pieces. It asked questions that helped him realize they were missing.

This is different from asking AI to generate a plan for you. You bring the plan. AI asks the questions a good coach would ask. The result in Caplan's case was a PDF with structured tables, but the value was in the dialogue that preceded it.

Tactic 5: Add structure to meandering thoughts

The final tactic treats AI as an organizing layer. You do the thinking. AI does the formatting. After a long voice note or written brainstorm, you ask AI to pull out the main themes, identify contradictions, or sort ideas into categories.

This works particularly well for synthesizing user research, organizing feedback from multiple stakeholders, or turning a messy spec into a structured document. The thinking remains yours. The structure becomes clearer.

The limitation worth noting

Caplan acknowledges a real constraint: without detailed context, AI responses trend toward bland generics. The sparring partner only pushes back effectively if you give it enough material to push against. A one-sentence prompt gets a one-dimensional response. The quality of AI-assisted thinking scales with the effort you put into framing the conversation.

Frequently Asked Questions

How do I get AI to actually challenge my ideas?

Prompt explicitly for pushback. Include phrases like 'play devil's advocate' and ask specific questions: What are the strongest counterarguments? What risks am I missing? Generic prompts yield generic validation.

What's bionic dictation?

Speaking your thoughts aloud and using AI to transcribe and organize them into structured text. It combines voice dictation with AI summarization to turn rambling into outlines.

Can I use AI for sensitive decisions without sending data to the cloud?

Yes. Offline AI tools like Jan run locally on your machine. You get the sparring partner functionality without exposing proprietary information to external servers.

Does AI thinking assistance actually improve decision quality?

It depends on how you use it. Asking AI to generate decisions for you offloads thinking. Asking it to question your decisions exercises thinking. The second approach surfaces blind spots before they become costly.

Source: Fast Company / Jeremy Caplan