A new article in MIT Sloan Management Review argues that most companies introduce technology in the wrong order. Managers are told to cut what annoys staff and add what stretches them, and the authors argue that the first job should be finished before the second one starts.
Two kinds of stress do not cancel out
Technology puts two kinds of load on people. One kind is an obstacle: the system that crashes, the notification that follows you home, the login that expires. The other is a challenge: learning a complex tool and becoming the person on the team who knows how to use it. It is natural to think the good kind makes up for the bad kind, so that enough interesting work will make people forgive the crashes.
It does not work that way. A person who gets both at once feels excited and fed up in the same hour, and holding two feelings that pull against each other is tiring in itself. Satisfaction drops and more people leave. It is like asking someone to enjoy a hike while a stone is in their boot: every hill ends up being blamed on the stone.
What the right order looks like in an AI rollout
Take the office of six that was asked to use AI on supplier invoices. Say each person loses 20 minutes a day to logging back in and closing pings. That is two hours a day for the team, or about 40 hours a month, a full working week. A half-day course on the AI assistant costs the same six people 24 hours. So fixing the logins first gives back more time than the training needs, and the training arrives into a week that already feels lighter.
Once the friction is gone, the new tool is the one demanding thing, and a manager can make it feel like a gain in skill instead of a burden. Let people choose which invoices to try it on, and pair each beginner with a colleague who is already good at it. That is closer to how people actually learn a new tool than a video course, and it gives the learner someone to ask when the tool disappoints them.
The first step costs nothing: ask the team for the three things in their current tools that waste the most time, and fix the cheapest before anyone is asked to learn something new. It also explains part of why AI can leave workers with a heavier load, not a lighter one: the new tool often arrives on top of everything that was already wrong.
Where the advice stops
The authors say their research did not cover generative AI, and the gap matters. Slow logins and pings are about how work gets done. An AI assistant also raises the question of whether the work will still need the person doing it. A team that suspects it is training its own replacement will see the hard new tool as a threat, not a challenge, and no amount of fixing logins changes that.
So the order has a step before the first one. Tell people plainly what the tool means for their roles, then fix what irritates them, and only then ask them to stretch.