What Happens to Your Product Decisions When AI Thinks First
- blogs, product management
- 4 min read
Author: Arnould Maren Joseph – Product Marketer
Over the past two years, product organizations have undergone a subtle but important shift.
Teams are operating at a pace that would have seemed unrealistic not very long ago. AI is now deeply embedded into everyday product work, helping generate PRDs, synthesize user interviews, map competitive landscapes, prepare presentations, and accelerate workflows that once consumed entire days.
The gains in speed are undeniable. Teams are shipping faster, documenting more thoroughly, and processing larger volumes of information than ever before.
At the same time, another pattern has started to emerge beneath the surface.
In conversations that require original thinking rather than generated output, many product managers find themselves relying heavily on what the tools suggest, rather than on a clearly formed point of view of their own. The challenge is rarely a lack of effort or capability. It is more subtle than that.
As AI begins handling larger portions of the cognitive workload, product teams are starting to confront an important question: What happens to judgment when the struggle required to build it is gradually removed from the process?
The research nobody in the PM world is talking about
In January 2025, a study published in the journal Societies examined 666 people across different age groups, educational backgrounds and levels of AI tool usage. The finding was direct: frequent reliance on AI tools showed a significant negative correlation with critical thinking abilities, mediated by cognitive offloading.
Cognitive offloading is the technical term for what happens when you stop holding a thought in your head and let a tool hold it for you. It is not new. We have been doing it with calculators, GPS and search engines for decades. The question has always been whether the offloading frees up mental capacity for harder thinking, or whether it gradually degrades the capacity it was supposed to free.
For AI tools used in writing and decision-making, a 2025 MIT study found something specific: individuals who consistently used LLMs for these tasks showed reduced brain activity, diminished memory retention and less original thinking. The more they depended on AI for cognitive tasks, the less engaged their own minds became. The researchers called it cognitive atrophy.
The Harvard Gazette covered this in November 2025, noting that MIT’s finding was small and not peer-reviewed, and still decided it was worth reporting. Their framing was careful: “It depends on how we engage with it, as a crutch or a tool for growth.” That distinction, crutch versus tool, is the thing worth sitting with.
Longitudinal studies are now finding something more unsettling. Sustained AI use does not just affect performance while the tool is being used. It alters underlying abilities. Writers who rely on AI assistance exhibit weaker recall of their own text and a diminished sense of ownership. Crucially, these deficits persisted even after access to the AI was withdrawn. The tool was gone, but the capacity it had been substituting for did not automatically return.
This is the part that should give product managers pause.
How PM intuition actually gets built
The research on expert intuition goes back to Gary Klein’s work in the 1980s and 1990s, studying how firefighters, military commanders and intensive care nurses made decisions under pressure. His finding, later popularised by Malcolm Gladwell in Blink, was that expert intuition is not a mystical sixth sense. It is pattern recognition. And patterns are built through struggle.
The brain processes ambiguous situations by scanning against previous experience, assigning what researchers call emotional shortcut tags to memories of outcomes. When you have seen a certain type of user complaint resolve in a certain way, when you have watched a feature built for the wrong reason fail in a predictable way, when you have sat in enough roadmap conversations to know which kind of stakeholder pressure leads where, you build a library of tagged patterns. That library is what product intuition actually is.
The Academy of Management Review defines intuition in managerial decision-making as affectively charged judgments that arise through rapid, nonconscious and holistic associations. The keyword is nonconscious. Intuition does not feel like thinking. It feels like knowing. But the knowledge only develops through repeated conscious engagement with difficult problems.
As Intercom’s research on product judgment puts it, no one starts out with strong product judgment. It is not innate. It takes years to build, and it ranges from very weak to very strong based on how deliberately it has been developed.
The mechanism for developing it is, specifically, wrestling with ambiguity before a resolution appears. Sitting with a problem long enough that your brain has to work. Forming a view before you see someone else’s view. Being wrong, noticing you were wrong, updating the pattern. Repeat several hundred times, and the result is what senior PMs call product sense.
Now consider what happens when AI generates the first draft.
The first draft is where the thinking happens
This is the part most people miss.
When a PM writes a problem statement from scratch, they are not just producing a document. They are forcing their thinking into a structure. The act of writing requires them to decide what the problem actually is, which user it belongs to, why it matters now and not later. These micro-decisions, many of them uncomfortable, are the cognitive workout that builds judgment.
When the AI generates the first draft, those micro-decisions get made before the PM engages. The structure is already there. The PM’s job becomes editing rather than thinking. And editing from an existing structure is a fundamentally different cognitive task from constructing one. It is the difference between following a recipe and cooking from ingredients.
The same thing happens with user research synthesis. When you read through forty interview transcripts yourself and have to decide what the pattern is, you engage with the contradictions, the outliers, the things that do not fit the narrative. That engagement is where instinct about users gets formed. When AI synthesises the transcripts and hands you a neat list of themes, you skip the contradictions. You see the conclusion without doing the reasoning. The output looks the same on the slide. The understanding is not.
Research on AI-assisted writing found that consistent users exhibit what the researchers called an illusion of competence. They mistake the fluency of AI-generated output for their own depth of understanding. The document reads well. The thinking behind it is shallower than it appears.
The particular risk for product managers
This risk is not evenly distributed. It is sharpest in the early and middle stages of a PM career, when the patterns that form product intuition are still being built.
A PM with ten years of experience has already made hundreds of product calls, seen the consequences, and adjusted their mental models. When they use AI to generate a first draft, they are editing against a rich internal library. They catch the things that are wrong. They know which suggestions to override and why.
A PM with two years of experience has a thinner library. When they edit an AI-generated problem statement, they are less able to notice what is missing because they have not yet built the pattern recognition to see it. They ship the document. The thinking did not happen. The pattern does not get added to the library.
Over time, this compounds. The senior PM using AI gets faster without losing much judgment. The junior PM using AI gets faster, while the judgment that should be developing does not. The gap between their outputs narrows. The gap between their actual capability widens in a direction nobody can see.
This is what the research on cognitive atrophy is pointing at. It is not that AI makes people stupid. It is that AI, used as a crutch rather than a tool, interrupts the specific kind of struggle that builds expertise. And expertise in product management is almost entirely about what happens in the struggle.
A distinction that matters
None of this is an argument against using AI. That argument is both wrong and pointless.
The distinction that matters is between using AI to extend your thinking and using AI to replace the thinking you have not yet done.
Using AI to extend thinking looks like this: you have formed a point of view on the problem, you have written your first draft, you use AI to pressure-test your logic, find counterarguments, and identify what you might be missing. The thinking happened first. AI makes thinking better.
Using AI to replace unformed thinking looks like this: you have a problem to frame, the blank page is uncomfortable, you ask AI to generate a first draft and work from there. The thinking never happened. The AI produced an output that looks like thinking.
The outputs can be hard to distinguish. The underlying cognitive development is completely different.
Gary Klein’s research found that expertise develops through what he called deliberate practice under conditions of uncertainty, with feedback on outcomes. The feedback loop is what calibrates the pattern library. When AI removes the uncertainty before the PM engages with it, the feedback loop does not fire. There is no cognitive event to calibrate.
What to actually do
Be deliberate about which cognitive tasks you protect from AI assistance.
The ones worth protecting are the ones that build the judgment you will need in five years. Forming a point of view on a problem before you read anyone else’s take. Writing the first paragraph of a problem statement before opening any tool. Sitting with user interview notes long enough to form your own read of what the pattern is before asking AI to summarise it.
These feel slower. They are slower. That is the point. The slowness is the cognitive workout. It is not inefficient. It is an investment.
Use AI aggressively for tasks that consume time without building judgment. Formatting, note-taking, research compilation, draft polishing, and generating options to evaluate. These are legitimate uses of a genuinely powerful tool. The issue is not AI; it is which tasks you are asking it to own.
The research on cognitive offloading has one genuinely hopeful finding: higher educational attainment served as a protective buffer against cognitive atrophy from AI use. The researchers believe this is because people with strong existing knowledge frameworks are better at critically evaluating AI output rather than accepting it. In other words, the richer your existing mental library, the less likely you are to have it displaced by the tool.
That is actually an argument for going slower in the early and middle stages of your PM career, not faster. Build the library first. Then use the tools to extend it.
The question nobody is asking
The conversation about AI and product management has mostly been about replacement. Will AI take the PM job? The answer is probably not, for the same reasons everyone cites: judgment, relationships, ambiguity, the call nobody else wants to make.
The question that is not being asked is subtler and more urgent. What happens to the judgment when the conditions that develop it are systematically removed?
The answer, based on what the research is showing, is that it develops more slowly. Or incompletely. Or in ways that are invisible until the tool is not available and the gap becomes apparent.
The PMs who will have genuine product sense in ten years are not the ones who used AI the most. They are the ones who were careful about which parts of the thinking they kept for themselves.
Frequently Asked Questions
1. How is AI changing product management decision-making?
AI is changing product management by speeding up tasks like documentation, research synthesis, and planning. At the same time, many PMs are becoming more dependent on AI-generated thinking, which can weaken independent judgment and original problem-solving over time.
2. Can AI reduce critical thinking skills in product managers?
Research is starting to show that heavy dependence on AI tools can reduce critical thinking and weaken memory retention. When PMs rely on AI to do the thinking before forming their own perspective, they may lose the cognitive struggle that helps build strong product judgment.
3. Why is product judgment important in the age of AI?
Product judgment matters because AI can generate answers quickly, but it cannot replace experience-based decision-making. Strong PM judgment comes from years of seeing product successes, failures, user behaviour, and difficult trade-offs play out in real situations.
4. Does using AI make product managers less creative?
AI can reduce originality when product managers depend on generated outputs too early in the thinking process. Many PMs unknowingly shift from creating ideas to editing AI suggestions, which changes how deeply they engage with the actual problem.
5. What is cognitive offloading in product management?
Cognitive offloading happens when people rely on tools to handle mental work they would normally do themselves. In product management, this can happen when AI writes first drafts, summarises research, or frames problems before the PM has formed their own understanding.
6. How should product managers use AI without weakening their thinking?
The healthiest way to use AI is to let it extend thinking rather than replace it. Strong PMs usually form their own perspective first, then use AI to challenge ideas, improve clarity, or spot gaps they may have missed.
