How to Keep Your Judgment Sharp in the AI Era

Ellie at Willa, Editorial lead, Willa · Published August 5, 2026 · Updated August 25, 2026

What if the biggest risk of AI isn't that it takes your job, but that it quietly takes over your thinking? That was the question at the heart of a recent live Willa session, "The Judgment Edge: Thinking Sharply When AI Does Everything Else," led by Crystal King, Senior Evangelist at HubSpot Academy. Crystal created HubSpot Academy's Critical Thinkers course for marketers and the free Critical Thinker's Guide to AI, holds a master's in critical and creative thinking, and has written four novels. Hosted by Willa founder Allison, the session dug into what research says AI use does to our brains, and the frameworks that keep your judgment intact while you keep using the tools.

Why is it so hard to think critically when using AI?

Crystal's starting point wasn't AI. It was the brain. "AI doesn't create new cognitive problems," she explained. "It actually exploits the ones that we already have."

The skill that underpins all critical thinking is metacognition: thinking about your own thinking. It's catching yourself mid-email and asking: am I being emotional or strategic? Most of us believe we're good at it, but Crystal cited organizational psychologist Tasha Eurich's finding that only 10 to 15% of people are actually self-aware. The rest of us spend much of the day on cognitive autopilot, the brain's default for handling too much information with too little time.

Autopilot is exactly what makes AI risky. When information is polished and well presented, it slides through your mental filters faster, and AI output is everything autopilot loves: fluent, confident, and instant. Crystal pointed to three biases AI exploits perfectly:

• Confirmation bias: you're more likely to accept AI output that confirms what you already believe.
• The halo effect: polished, professional-sounding text reads as credible even when it's wrong.
• Framing bias: AI presents everything with the same confident, authoritative tone. There's no visual signal that says "I'm guessing," and, as Crystal noted, it guesses all the time.

Put those together and "the output looks right, it sounds right, and it confirms what you thought." A perfect storm. She also cited Harvard Project Zero's seven "thinking dispositions," warning that the three most at risk from AI use are being intellectually careful, seeking and evaluating reasons, and metacognition itself. These are habits you can strengthen or let atrophy.

When does AI actually make your work worse?

In 2023, CNET used AI to write financial explainer articles, with professional editors reviewing every piece. Out of 77 published articles, 41 needed substantial corrections. One even got compound interest math wrong, advice that would have cost readers money. The editors weren't careless. The writing looked good, so their brains processed it as credible. Crystal called this processing fluency: confusing readability with truth. "It doesn't mean it is awesome," she said of AI output. "It just looks and feels awesome."

The deeper problem is knowing where AI helps and where it hurts. Crystal described a Harvard Business School field experiment with 758 BCG consultants. Inside what researchers called the safe zone (pattern matching, drafting, summarizing), AI users worked faster and produced up to 40% better work. But on tasks requiring nuanced judgment, multi-step reasoning, or contextual strategy, consultants using AI performed 19 percentage points worse than those who didn't use it at all. Worse than nothing.

And the boundary between those zones is uneven. AI won't tell you when you've crossed it. When it goes wrong, asking "why did you hallucinate?" is a dead end: "You can't debug something that's working as intended," Crystal said. AI assembles patterns into the most plausible-looking answer, not necessarily the correct one. The productive question is about yourself: what made me accept this answer?

What is cognitive surrender, and am I doing it?

Crystal highlighted a Wharton School distinction between two ways of handing work to a machine. Cognitive offloading is fine: it's what you do with a calculator. You delegate a task but still own the problem: you choose the approach and evaluate the result. Cognitive surrender is different: adopting AI's output as your own answer without meaningfully engaging with it. You're not delegating a subtask. You're handing over the thinking itself.

The shift is gradual: you edit the draft, then edit less, then stop editing because it's easier. And it shows up in the brain. Crystal described an MIT Media Lab study that put 54 people in an EEG lab to write essays with ChatGPT, a search engine, or nothing. The ChatGPT group showed the weakest brain connectivity while writing, and when the AI was taken away, they couldn't remember their own earlier work. Researchers call it cognitive debt, and the neural differences appeared within a single writing session. A larger study of 666 participants she cited found that heavier AI use correlates with a decline in the inclination to think deeply. "Not the ability to think deeply, but the inclination," Crystal emphasized. "You can still do it. You just stop wanting to."

There are professional costs, too. Crystal cited a Workday survey of 3,200 professionals in which 85% said AI saves them time, but 37% of that saved time gets eaten by rework: for every ten hours gained, almost four go to cleanup. And she pointed to Stanford economist Erik Brynjolfsson's analysis of payroll data covering 25 million workers: workers aged 22 to 25 in the jobs most exposed to AI have seen a 13% relative decline in employment since 2022, while older workers in the same roles are stable or growing. Entry-level work (the research, the analysis, the first drafts) is where professionals traditionally learn to build judgment, and it's being hollowed out.

How do I know which tasks are safe to hand to AI?

Crystal described what researcher Madeleine Clare Elish calls the moral crumple zone: when an automated system fails, responsibility rolls downhill to the nearest human. If you set the AI in motion, you take the blame, even when the AI got it wrong.

Her framework for avoiding that starts with two axes: how much nuance does the task need, and how big is the impact of an error? Low-nuance, low-stakes work (internal brainstorms, reformatting data, a first draft you'll completely rework): let AI run. High-nuance, high-stakes work (client deliverables, published content, strategy recommendations, anything with your name on it) is deep human ownership territory that needs line-by-line review. "You can't verify nothing, and you can't verify everything," she said. The skill is classifying which is which.

Before delegating, she suggested four questions:

1. Would I know if it's wrong? If you lack the domain expertise to catch a hallucination, you can't fully delegate the task.
2. What's the cost of an error? Internal notes can be skimmed; published financial data cannot.
3. Am I asking AI to generate options or to make a judgment? "AI is awesome at expanding your options," Crystal said, "but it's not the best at making the final call."
4. Does consistency matter? For compliance language, brand voice, or legal terms, AI's natural variability works against you.

For verifying output, she teaches the SIFT method, created by researcher Mike Caulfield: Stop before accepting anything at face value, because AI output has no author, no editor, no peer review. Investigate whether a claim is generated or verified. Find secondary confirmation from a human source. Trace claims back to the original study, not an article about an article about it. When AI research for one of Crystal's own courses surfaced citations it couldn't source, she cut them. She also routinely pits Claude, Gemini, and ChatGPT against each other to fact-check the same material.

What daily habits keep your critical thinking sharp?

Crystal closed with four habits to build:

1. Keep doing hard thinking on purpose. Each week, pick one task you could hand to AI and do it yourself instead. Keep the muscles active.
2. Practice metacognition about your AI use. A few times a week, ask: what did I hand off without thinking? Where do I feel the pull to skip the thinking and just accept?
3. Protect your curiosity from the convenience trap. Before consulting AI on anything involving judgment or strategy, write down your own three ideas first, then compare. "If you don't put your own idea on the table first, the AI's answer wins by default."
4. Treat verification as a thinking practice, not a chore. Before anything high-stakes ships, ask: if this turns out to be wrong, can I explain why I believed it was right? If you can't answer that, you haven't really reviewed it.

None of this means quitting AI. As Allison put it, going cold turkey isn't realistic: the tools are too valuable. "But we can't blindly trust these tools," she said. "We always need to be questioning, and thinking critically." Crystal ended with Stoic philosopher Epictetus: "No man is free who is not master of himself." True freedom means thinking for yourself.

Key takeaways

• AI doesn't create new cognitive problems: it exploits existing ones. Confirmation bias, the halo effect, and framing bias make polished AI output feel true whether or not it is.
• There's a line between cognitive offloading (delegating a task while owning the problem) and cognitive surrender (adopting AI's answer without engaging). The slide from one to the other is gradual, and it starts affecting your brain within a single session.
• AI helps most on pattern matching, drafting, and summarizing. In one study Crystal cited, consultants using AI on nuanced judgment tasks performed 19 points worse than those using none.
• Match scrutiny to stakes: sort tasks by nuance and impact of error, and give anything with your name on it line-by-line review. When AI fails, accountability rolls downhill to you.
• Keep the muscles working: do one hard thinking task yourself each week, write your own ideas down before asking AI, and never ship what you couldn't defend.

Critical thinking is itself an AI skill, arguably the one that makes all the others valuable. If you're building your toolkit, start with AI Skills for Non-Technical Professionals: Where to Start. Judgment about when and how to use AI is also exactly the kind of capability worth putting on your resume.

Join us for the next one

This conversation came from a live Willa event, where members traded verification tricks, fears, and hard-won lessons in real time. Willa hosts expert-led sessions like this almost every week for women building careers with AI. See what's coming up on our events page, or become a member and join a community that's learning to use AI, and stay sharp, together.

• critical thinking
• AI at work
• cognitive offloading
• AI literacy
• professional development

Join Willa, the community where women learn AI together.

All posts