AI and Critical Thinking: 7 Levers Any Leader Can Pull

 One of the biggest fears executives have about AI is that their people will stop thinking. The analyst who used to build the model from scratch just asks Copilot. The manager who used to wrestle with a tough call takes the first answer that sounds good. A few years from now, nobody remembers how to do the work without it.

That fear is real, and honestly, it can be founded. Think about GPS. Most of us used to read maps and watch for exit signs, and now plenty of us couldn't find our way home from the grocery store if our phone died (I'm in this group, no shame). GPS didn't take anyone's sense of direction away. We just stopped using it.

The research suggests AI can do the same thing to critical thinking. Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers who use generative AI at least once a week. People with more confidence in AI reported putting less effort into critical thinking, and people with more confidence in their own skills reported putting in more. The study is self-reported, but the pattern is clear: the people most likely to accept a bad AI answer are the ones who feel least sure of their own expertise.

So yes, the fear has some ground to stand on. But what happens next is all in how we approach AI.

How we teach AI decides what happens to critical thinking

Most AI training teaches people how to use the tool: write a prompt, summarize a meeting, draft an email. That matters, and it's where adoption starts (I've led that kind of rollout at scale, 77,000 users, 93% adoption, in 2 months).

But don't just teach people how to use AI. Teach them HOW to use it.

There's a big difference between these two prompts:

• "Write me a recommendation for next year's budget."

• "Here's my recommendation for next year's budget and why. Poke holes in it, tell me what I'm missing, and ask me the questions my CFO will ask."

The first one hands the thinking to AI. The second one uses AI as a thought partner, so the person does the thinking and AI pushes it further. Same tool, same license, very different outcome for AI and critical thinking.

It's also worth saying that most people aren't skipping the thinking on purpose. They're busy, the answer looks polished, and they may not know they need to check their responses.  

Seven levers to protect critical thinking with AI

You don't need a company-wide initiative to start. Every lever below works on a single team, so if you lead people, you can pull any of them this month, whether or not anyone above you is talking about this yet. If you're an executive, each one also has a way to take it org-wide.

Lever 1: Say the expectation out loud

People follow what their leader says and rewards. If the only message they hear is "use AI more," they hear "go faster."

On your team

At your next team meeting, say plainly that you expect people to use AI and to check it. Describe what good looks like: "Use Copilot for the first draft, then tell me what you'd change and why." Share a time you questioned an AI answer yourself.

Org-wide:

Put the same message in a town hall or all-hands email, in your own words. 

Lever 2: Teach AI as a thought partner

Most people have only seen AI used to produce things. Few have seen it used to challenge their thinking.

On your team: 

Share three thought-partner prompts and use one live in a meeting: "Poke holes in this," "What assumptions am I making that are most likely wrong?" and "Ask me questions one at a time to help me work through this." Save them somewhere the team can find them.

Org-wide: 

Add thought-partner prompting to your AI training, right next to the drafting and summarizing.

Lever 3: Think first, then prompt

When AI writes the first draft, people react to it instead of forming their own view.

On your team: 

On important work, ask people to jot down two or three bullets of their own take before opening Copilot, then compare. Try it together once so it feels normal.

Org-wide: 

Add a short "my recommendation and why" section to decision memos and business cases, written before any AI analysis.

Lever 4: Name the human-owned decisions

Over-trust grows in the gray areas. When nobody has said which decisions need a human, people guess.

On your team: 

Make a one-page list of the decisions on your team that always need human judgment and sign-off. A sales team's list looks different from a finance team's, so build it with your people.

Org-wide: 

Publish a company list covering the big ones, like hiring and performance, customer commitments, financial reporting, and policy interpretations.

Lever 5: Make the reasoning visible

If nobody can see the check, nobody can coach it or recognize it.

On your team: 

Ask for a two-line "AI use note" on key deliverables covering what AI drafted and what the person checked or changed. Treat it like citing a source, and thank people for good catches.

Org-wide: 

Make the note a standard for deliverables that go to senior leaders or clients.

Lever 6: Protect early-career skill building

The Microsoft and Carnegie Mellon study linked confidence in your own skills to more critical thinking. New hires haven't had time to build that confidence yet.

On your team: 

Have new team members draft the first version of core work (an analysis, a client summary, a project plan) on their own, then use AI to improve it. You learn the roads before you lean on the GPS.

Org-wide: 

Ask HR and L&D to identify the core skills each role should learn by hand first and build them into onboarding.

Lever 7: Coach and measure the thinking

Adoption dashboards count prompts. They can't tell you whether anyone is thinking.

On your team: 

Swap "Is it done?" for "What did you push back on?" in your next three 1:1s. Once a month, review five AI-assisted deliverables with your team for accuracy and reasoning, and recognize the people who caught AI mistakes.

Org-wide: 

Add judgment to competency models and pair usage metrics with a quarterly quality sample. Skip the usage leaderboards (bad idea, but that's a whole other post).

Try This: Ask your team one question (15 minutes)

At your next team meeting, ask:

"Where in your work would you accept an AI answer without checking it?"

Write the answers on a whiteboard or in a shared doc, then sort them into three groups: fine to trust (low stakes, errors are easy to spot), needs a quick check, and needs human judgment.

The "needs human judgment" group becomes your team's human-owned decisions list from Lever 4. The "needs a quick check" group is where thought-partner prompts from Lever 2 earn their keep. Post both on your team SharePoint page and come back to them in 30 days to see what's changed. It needs to be normalized, and that takes some time.

GPS didn't wipe out anyone's sense of direction overnight. It happened one "recalculating" at a time (the most judgmental word in the English language), because nobody had a reason to look up from the screen. AI and critical thinking can grow together when leaders give people that reason, one team at a time.

Curious- Which lever could you start with your team this month?