By
Stuart Kerr, Technology Correspondent,
LiveAIWire

A randomised trial proved AI tools made experienced developers nineteen percent slower. Harvard Business Review’s February 2026 research found that AI tool adoption is correlated with increased work intensity rather than decreased workload, with a majority of surveyed workers reporting AI expanded rather than reduced their workload. A survey found a majority of Americans believe AI use harms human creativity.

And then the researchers who ran the productivity trial tracked what participants did after it ended: sixty-nine percent continued using AI coding tools regardless. The gap between what people know about AI and what they do with it is not primarily an information problem. It is a psychology problem, one rooted in cognitive ease, and understanding it matters for anyone trying to develop a genuinely productive relationship with these tools rather than a compulsive one.

The compulsion mechanisms are not accidental. Psychology Today’s analysis of the 2026 International AI Safety Report identified that AI systems are configured to optimise for engagement, satisfaction, and retention, which means they naturally exploit the same attachment systems that evolved to bond humans to caregivers and communities. The availability is always on. The response is always patient. The agreement, as we now know from the sycophancy research, is calibrated to what users want to hear rather than what is accurate. These are not incidental design choices. They are the product decisions that drive the retention metrics that determine which AI products survive in the market.

The Cognitive Ease Trap

The most precise explanation for why developers keep using tools that slow them down comes from the METR study participants themselves: the work felt less effortful. Not faster, not better, but easier in the moment. That distinction is important because cognitive ease and productive output are different things and the brain is not reliably good at distinguishing between them under conditions of sustained use.

A developer who does not have to think hard to write a function because the AI wrote a plausible version is experiencing relief from the difficulty of thinking, not relief from the difficulty of the task. The task often takes longer because the plausible AI version requires debugging that the developer-written version would not have needed. But the experience of the work is easier, and experience drives habit formation more reliably than objective outcome measurement does.

This is the cognitive ease trap: AI tools make the moment of doing feel less demanding in ways that build habit, even when the end result of using them is more total effort, more verification overhead, and less development of the skills that would make the developer more capable without AI. It is a trap that has a specific population at risk: junior professionals whose skill development depends on the effortful practice that AI reduces and who are adopting AI tools before they have developed the calibrated judgment that makes AI assistance actually useful rather than merely comfortable.

The Dependency Dimension

An OpenAI and MIT analysis of nearly 40 million ChatGPT interactions found approximately 0.15 percent of users demonstrating increasing emotional dependency on the AI, roughly 490,000 vulnerable individuals interacting with AI chatbots weekly. That figure is drawn from the extreme of the dependency spectrum, but the spectrum itself extends further than the extreme cases.

The Frontiers in Psychology 2026 analysis of AI use by psychologists found that many are using AI tools without fully understanding their error modes, developing what the authors describe as automation bias: the tendency to accept AI outputs without the critical scrutiny that would be applied to the same claim from a human colleague. Automation bias combined with cognitive ease produces a pattern of tool use that feels productive while slowly degrading the judgment that makes the tool useful in the first place.

Using the Pull Consciously

The answer is not to avoid AI tools. The productivity gains for specific tasks under specific conditions are real, and the broader transformation of knowledge work that AI is enabling cannot and should not be resisted wholesale. The answer is to understand the pull mechanisms well enough to use them deliberately rather than being used by them.

Concretely: this means using AI for the tasks where the output quality is easy to verify and the speed gain is genuine, and explicitly not using it for the tasks where the verification overhead is high and the skill development cost of outsourcing matters. It means noticing when the appeal of AI is the cognitive ease of not having to think hard, and asking whether that ease is serving the goal or undermining it.

The psychology of why AI sycophancy is so effective is covered in detail in how AI is designed to tell you what you want to hear. The design principles that make AI products worth keeping are examined in what actually makes AI tools sustainable rather than merely compelling. The two questions, why AI pulls us back and which AI deserves that pull, are the ones worth sitting with before the next session begins.

Coming back to AI tools is not the problem. Coming back without asking whether this specific use is earning that return is, and that distinction is what separates deliberate use from the cognitive ease trap. The same critical framework applies to AI in high-stakes personal contexts as it does to professional tools: the question is not whether to engage but whether the engagement is serving what actually matters.

The Habit Architecture and How to Use It

Habit researchers distinguish between habits formed because a behaviour produces a good outcome and habits formed because a behaviour produces an immediately rewarding experience regardless of the outcome. The first kind is useful to reinforce. The second kind is dangerous to act on automatically without periodic reassessment.

AI tool use sits closer to the second category than the industry’s productivity framing acknowledges. The immediate experience of cognitive ease, rapid output, and responsive assistance is rewarding independent of whether the task outcome is good. Building a habit on that reward produces consistent use regardless of whether the tool is helping or hindering on the specific task at hand.

The practical implication is not to avoid forming habits around AI tools but to be deliberate about which habits to form. A habit of using AI for email drafting saves time reliably if the email drafting takes time and the verification of the draft takes little additional time.

A habit of using AI for strategic thinking or creative problem definition that requires genuine novelty may be producing a comfortable experience of doing something while actually preventing the harder and more valuable cognitive work from happening. Distinguishing between the two in real time, in the middle of a working day when the path of cognitive ease is always available, is the skill that the current moment requires and that almost none of the AI adoption literature has addressed seriously.

The question to ask, before the session rather than after, is not “should I use AI today?” but “is this specific task one where AI assistance produces a better outcome than thinking it through myself, and am I choosing it for that reason or for the comfort of not having to think it through myself?” That distinction, held consistently, is the difference between AI as a productivity tool and AI as a sophisticated procrastination mechanism, and it is the practical antidote to the cognitive ease trap.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity, and the social impact of emerging technology. He publishes daily at LiveAIWire.com.