In March 2026, I ran a session for the Institute of Executive Coaching and Leadership (IECL) on neuroplasticity and coaching. That phrase "an old dog can't learn new tricks" is often stuck in people's minds, but it's simply not true. The brain's capacity to rewire itself based on experience, attention and repeated action is genuinely good news.
However, there is a flip side that doesn't get talked about nearly enough.
Neuroplasticity works in both directions
And that means that how you use AI today will impact what your brain is capable of next year.
What London taxi drivers taught us about the brain
A particularly striking study followed London taxi drivers who had completed The Knowledge, the notoriously demanding process of memorising over 26,000 streets and thousands of points of interest across London. Researchers at UCL (led by Eleanor Maguire in 2000 and 2006) found that these drivers had measurably larger posterior hippocampi, the region associated with spatial navigation and memory, than people who weren't taxi drivers. It was also larger than in bus drivers, who follow pre-determined routes. The longer they had been a taxi driver, the more pronounced the difference.
The brain had literally reorganised itself around what it was repeatedly asked to do.
When people rely on GPS, the effect looks to run the other way. Dahmani and Bohbot (2020) found that heavier lifetime GPS use went with worse spatial memory when people had to navigate for themselves. Then, in a follow-up three years later, the people whose GPS use went up the most showed the steepest decline. (This was behavioural work, not brain imaging, and the follow-up group was small, so this is directional, not conclusive.)
The brain is deeply practical. It allocates resources to what you repeatedly do and withdraws them from what you outsource.
Which brings us to AI. And in this post, by AI, I'm referring to LLMs: large language models like Claude and ChatGPT.
The early signals
We are at the beginning of the research on AI and cognitive effects, and I want to be careful not to overstate what we know. But the directional signals are worth paying attention to.
A peer-reviewed study from Microsoft Research and Carnegie Mellon University (Lee and others, 2025), involving 319 knowledge workers across 936 real-world AI use cases, found that higher confidence in AI correlated with less critical thinking, not more. The less people trusted themselves, the more they deferred to the AI, regardless of the quality of what it produced.
A study from MIT Media Lab (Kosmyna and others, 2025) found that participants writing with ChatGPT showed markedly weaker brain connectivity than those writing without it. When the AI-assisted group then switched to unassisted writing, they showed weaker connectivity than participants who had never used AI at all. The researchers called this pattern "cognitive debt". (When I first wrote about it, this study hadn't been peer-reviewed, and the sample is small, only 54 people, so it's not conclusive.)
And a randomised experiment with 354 participants (Liu and others, 2026) found that AI assistance improved in-task performance, but that improvement did not transfer to unassisted work. People performed worse when the AI was removed than if they had never used it at all. They were also less persistent, which is key to building skills over time.
Reduced persistence is a general consequence of AI-assisted problem solving. Liu and others, 2026
Persistence is a foundational part of learning. Pushing through discomfort and frustration as you figure something out is exactly what strengthens neural pathways, as well as building your determination for future challenges.
How you choose to use AI matters neurologically, not just practically
During the IECL session, one of the questions we covered was: what is the difference, for your brain, between someone suggesting a solution to you and you arriving at that solution yourself?
The answer is significant. When you generate an answer yourself, you're more likely to remember it. Researchers call this the generation effect, and it's one of the more reliable findings in memory research. There's also early evidence that arriving at your own insight engages the brain's reward system, which may help reinforce it. Clean language in coaching allows insights to emerge from the client's own awareness. This is much more powerful than having a solution suggested to you (there are several non-neurological reasons for this too).
AI is very effective at creating suggestions that look good at first glance. And if you use it too early in your thinking process, you are short-circuiting exactly the cognitive effort that builds your judgement, expertise and capacity over time. That cognitive effort over time - that persistence - is essential to beneficial neuroplasticity.
Your brain continually revises and remodels, improving or slowly declining, as a function of how we use it. Michael Merzenich
What I've noticed when using AI
I can find AI genuinely addictive, particularly when building AI agents. You can create so much in such a short period of time. The pace of output is intoxicating. But if you're not careful, it's not always getting you closer to the outcomes you need.
I've also noticed how often it misleads you. AI is very confident in technical domains. It tends to gloss over the bugs it introduces and doesn't always surface what it's traded off: you usually need to ask. Build a website quickly and you might not realise the major accessibility issues until much later. Fix one bug and there's another. Fix that one and there's another. (And that's before we even get into AI "hallucinations", which are an inherent feature of how these systems are trained.) Meanwhile you've forgotten to eat, and your dog is ignoring you because you've ignored her all morning (sorry Wendy).
This pattern has a name in the research: work intensification.
AI consistently intensifies work rather than reducing it
This is partly driven by intrinsic reward: AI makes doing more feel accessible and rewarding, so you do more. Quite quickly, expectations adjust upward, workload expands, and individual workers take on more work (Ranganathan and Ye, Harvard Business Review, 2026). Over time, this can create cognitive strain. Cognitive impairment is one of four parts of burnout in the Burnout Assessment Tool, developed by Wilmar Schaufeli and colleagues at KU Leuven in 2020. The other three are exhaustion, emotional impairment and mental distance: a cynicism or lack of care that didn't exist before.
Then, when you look at AI use in teams, AI can increase work volume while reducing quality. Reviewing low-quality work doesn't save much time, and it's also less rewarding than starting from scratch, or building a skill with another human through feedback and experimentation.
So what do you actually do about it?
Here are some options at the individual level. The team challenge is a whole other conversation.
Before you default to AI, do your own thinking first. Jot bullet points, mind map, or talk it through with a colleague. Even five minutes of unassisted thinking before you engage AI means you are shaping the output rather than being shaped by it. This protects (some of) the neural work.
Talk to a human briefly. Not to get the answer, but to articulate your thinking. Verbalising to another person activates different cognitive processes than typing into a prompt box.
Ask AI to interview you, not answer you. Prompt it to use the Socratic method, or to ask you clarifying questions before it offers anything.
Timebox your sessions and set a physical alarm. Not a phone notification you'll ignore. An actual alarm that requires you to get up. It's similar to breaking other compulsive behaviour loops.
Check your recall. Once you have finished an AI-assisted session, close it and write down from memory what you decided, learned or produced. This reconsolidates the thinking and will help with recall later. It can be uncomfortable, and that discomfort is part of the learning process. Shifting from typing to writing or drawing can often spark new connections or ideas as well.
Ask whether AI will actually save you time. Not every task will benefit from using AI, and if the time saving is small, you're probably better off avoiding it and using your brain instead.
The upside is real, if you design for it
The London taxi drivers did not develop larger hippocampi by accident. They developed them because they spent years actively exercising spatial memory under pressure. The same principle applies here.
If you use AI as a thinking partner that challenges and extends your reasoning, your cognitive capacity can grow. If you use it as a replacement for thinking, the research suggests the opposite starts to occur over time.
We are all running a live experiment on our own brains right now. It is worth being intentional about the outcome you are designing for.
Related: Is AI resetting your social expectations?, on what agreeable AI does to our relationships.
