There is a pattern I have been watching play out at companies that adopted AI tools aggressively in 2023 and 2024. The early enthusiasm has given way to something more complicated. Employees are still using the tools but with less energy and less trust than they had initially. When you dig into it, the cause is usually the same: too many tools, too many promises, too much inconsistency between what the tools claimed to do and what they actually do in daily work.
I am calling it AI tool fatigue and I think it is worth naming directly because the response to it is different from the response to normal technology adoption friction.
Normal adoption friction is about learning curve and habit change. Employees need time to build new workflows and the friction decreases as the habits form. The appropriate response is support, time, and patience.
AI tool fatigue is different. It is not that employees have not had enough time to adjust. It is that they have adjusted, used the tools, and concluded that the tools are less reliable, less consistent, and less useful than they were presented as being. The appropriate response is not more time and patience. It is honest evaluation of which tools are delivering real value and which ones are consuming attention without returning it.
The companies I have seen handle this well made a counterintuitive move: they reduced the number of AI tools rather than adding more. They picked the two or three tools that were demonstrably delivering value for their specific workflows, invested in making those tools excellent, and explicitly retired the others. The employees who had been spreading their attention across six mediocre AI experiences consolidated onto two good ones and their engagement improved.
The companies that are handling it poorly are still in the mode of adding tools in response to capability gaps. Employee reports that the current AI assistant is not good at coding queries lead to a new AI coding tool. Employee reports that the knowledge base AI is unreliable lead to a new knowledge base AI. The tool count grows and the fatigue deepens.
The evaluation question I would encourage anyone managing an AI tool portfolio to ask once a quarter is simple: if we could only keep three AI tools, which three would we keep and why? The answer to that question usually reveals which tools are genuinely load-bearing and which ones are still around because nobody has made the decision to remove them.
The tools that survive that exercise tend to have a few things in common. They do a specific thing consistently well rather than many things inconsistently. Their failure modes are predictable enough that employees have calibrated their trust appropriately. They have improved over time rather than staying static. And employees mention them proactively when talking about how they work rather than only when asked.
The tools that get cut tend to be the ones that were adopted because they were impressive in a demo, useful for a specific project that has since ended, or added by someone who has since left the organization.
AI tool portfolios accumulate in one direction only. Decisions to add tools happen constantly. Decisions to remove tools require deliberate effort and political will. The portfolios that work are the ones where removal decisions happen as regularly as addition decisions.
Top comments (0)