One company in ten has a concrete plan for the skill erosion executives report

An employer that wants its staff using AI ^[11] can count adoption soon after a rollout. The four skills BCG says are fading fastest ^[3] show up later, in how a problem gets framed or how a proposed solution gets judged. I think that gap in timing explains part of the distance between the half of executives who report erosion ^[2] and the nine companies in ten without a concrete plan ^[15].

A skeptic would say that a survey of just over 70 executives ^[2] measures impressions. That is fair. Business Insider ties the erosion to heavy AI use ^[11], but the study as reported records what leaders see inside their own organizations, and the account does not describe any test of workers' skills. The answer is that those impressions belong to the people who would sponsor a plan. A third of companies have not discussed deskilling at all ^[5].

Martin's examples place the erosion in ordinary work. He said it is obvious when someone has had AI summarise a deck they were meant to read: they cannot answer follow-up questions ^[6]. "It's like they only know what the AI told them," Martin said ^[13]. He said he would not discourage workers from using AI for core tasks, but cautioned against accepting its output without scrutiny ^[12]. "Assume that what you see is wrong or misinformed, and then play the other side of it," he said. "Build a red team against what the AI says." ^[8]

Martin's other remedy has a cost that can be counted. A one-week break from AI every quarter ^[9] comes to four weeks a year of working without the tool ^[16]. Martin said that, given AI's promise of large productivity gains, companies have a vested interest in balancing automated tasks with work that does not involve the technology ^[10]. Business Insider's account describes the current position: with most organizations not stepping in, it is left to employees to protect their own skills ^[11].

The decision this quarter is how hard to push adoption and what to measure beside it. A company that sets usage goals with no skills plan will learn about erosion later, from the work itself, in follow-up questions that go unanswered. Martin's test asks whether a person can do the job at or above their usual level for a week without AI ^[9]. Run across a team, it would give leaders a measure of their own staff in place of the perceptions BCG collected ^[2]. "It can be dangerous if you rely on it too much," Martin said ^[14].