LLMs are Performance-Enhancing Drugs for the Mind

llm ai work

LLMs have transformed the world of work…yadda yadda yadda. Yawn.

What do you care about that? You’ve got a job to do. A mortgage to pay. Sometimes you wake up in the night and really need a wee but find it impossible to drag yourself out of your cosy little cocoon.

However, I am going to have to briefly touch on LLMs and transformation and all that gumph because a couple of interesting bits of research have been published that could end up suggesting a profound new future.

Ultimately, I’m going to argue that LLMs (or AI, if you’re being trendy) are PEDs (peformance-enhancing drugs) for the mind and that given this, we should think very carefully about how we want to use them.

Let’s arrive at this methodologically…

AI enables knowledge workers to be more effective

There’s now a large (and growing) body of evidence that AI enables knowledge workers to be more efficient and effective:

  • A 2024 study of 4,867 software developers at 3 different companies found that those using GitHub Copilot were significantly more productive, as measured through tasks completed (+26.08%), code commits (+13.55%), and code compiles (+38.38%). The study also found that inexperienced developers showed greater productivity increases.

  • A 2025 study of 5,172 customer-support agents found that access to generative AI-based conversational assistants increased productivity (as measured by issues resolved per hour) by 14%. Again, it was found that the least skilled workers benefitted most from the AI assistants.

  • A 2023 study of 758 consultants found that those who had access to AI tools completed 12.2% more tasks, approximately 25.1% more quickly, and with 40% higher quality than those who didn’t have access to AI tools. Again, the study found that those of lower skill level saw the biggest improvement in their output.

More recent studies have muddied the picture slightly; a 2025 study with 16 experienced developers found self-reported productivity improved but actual productivity fell. While the picture may therefore be more nuanced and depend on the developer, the tools and the problem at hand, it appears that any company not using AI to support its knowledge workers is likely to fall behind in the short-term.

Long-term AI use could negatively impact cognitive ability

  • A 2025 preprint of 54 people found that those using LLMs for essay writing used less cognitive effort, and showed worse performance over a 4 month period. Those using LLMs felt like they weren’t the owners of their essays, and struggled to recall quotes from the essays that they had just ‘written’.

  • A 2025 study of 319 people found that confidence in one’s ability to perform a task influenced the amount of critical thinking performed when using an LLM. People stated that using LLMs led to a reduction in critical thinking and an over-reliance on the LLM.

  • A 2025 study of 666 people found that frequent LLM use was strongly correlated with reduced critical thinking ability. People who relied more on AI said they delegated memory, reasoning, and decision-making to the LLM.

Though there are many fewer studies in this area, the initial signs are that LLM-boosted productivity may not be a free lunch and may have a real cognitive cost.

What that means for organisations

The large productivity gains enabled by Generative AI are already transforming hiring and retention practices across the tech industry. More than 150,000 people were laid off in the tech industry in 2024, and estimates suggest that those trends have only accelerated throughout 2025. Across industries, hiring is slowing in response to increased AI-boosted productivity. As knowledge workers see productivity gains, employers are finding that they can achieve the same things with fewer, AI-supported people.

However, hints (and hints are all we have at the moment) that long-term AI use could negatively impact the cognitive output of their staff should be a cause for concern. “Investing in people”, “training, learning and development”, and “our people are our biggest strength” have long been the mantra for employers of all types.

For organisations, the cost of AI use might be a long-term reduction in the quality of their workforce.

What that means for employees

Knowledge workers are characterised by both their knowledge, and their skill at applying that knowledge. These skills are improved through practice. If generative AI causes workers to spend less time and effort thinking about problems, it is reasonable to assume that those skills may atrophy.

Using AI may be good for your company right now, and being able to successfully use AI likely demonstrates that you will be able to operate in a skilful and efficient manner. Now. But if you spend less time and effort thinking about the problems that previously kept your skills sharp, you may well find those skills disappearing.

Performance-enhancing drugs

By now, hopefully, the analogy to performance-enhancing drugs becomes clear. Nobody denies that PEDs offer substantial performance gains for elite athletes. However, there is ample evidence that while they offer short-term gains, they contribute to the long-term health decline of whoever takes them.

Should coaches be able to prescribe that their athletes must use performance-enhancing drugs, for their short-term gain at the long-term cost of their athletes? I think we’re all pretty much agreed at this point that the answer is no: sounds sketchy AF.

Should companies be able to prescribe that their employees use cognitive-enhancing tools, for their short-term gain at the long-term cost of their employees’ cognitive ability? For now, while the evidence on the cognitive effect of AI use remains sparse, it would appear the answer is yes. But if it transpires that the early seeds of research showing the cognitive decline of those who use AI ultimately bloom, generative AI may find itself under yet more scrutiny.

What do I think you should do about this?

Honestly - I don’t know.

LLMs are a wonderful and powerful tool that can make you feel like a superhero - blasting through tasks that previously would have taken weeks in a matter of hours. Not learning how to use LLMs effectively feels like a terrible career (and life) move.

However, it appears that generative AI allows us to not think in the same depth and distances us mentally from our work. The long-term impacts of this aren’t currently understood but initial research indicates that LLMs might become a mental crutch that we begin to rely upon.

My advice: be thoughtful about where you want to be thoughtful. This blog post (like all my others) is written without LLM-assistance, because I believe writing is a skill worth having. Ignoring the LLM revolution that is underway feels like a misguided bet against technological progress. Ceding your cognitive function to a tool feels like a misguided bet against yourself.