The AI Myth: Why We’re Getting It Wrong Again

The latest iteration of technological disruption has brought with it a familiar narrative: AI is here to streamline our lives, and in doing so, cut costs by eliminating jobs. However, as Sam Altman pointed out earlier this year, some companies are using AI as a convenient excuse for workforce cuts they would have made regardless.

This “transformation” story typically goes like this: AI is here, headcount is a cost, and moving fast on both is what leadership looks like. But the data paints a different picture. When measured at the task level rather than the job level, AI’s impact changes dramatically.

Research by Anthropic’s team offers one of the most rigorous early attempts to measure AI’s labor market effects. Their findings suggest that even in occupations with high AI exposure – computer programmers, customer service representatives, and financial analysts – there has been no statistically significant increase in unemployment since ChatGPT launched. Cornerstone’s workforce intelligence platform reinforces this from a different lens. Tracking over 55,000 distinct skills across 1.3 billion job postings and 1 billion resumes globally, their data shows positive demand growth across 15 of 16 occupational categories regardless of AI exposure level.

What’s striking about these findings is the distinction between tasks and jobs. AI is primarily eliminating routine synthesis work, not entire roles. The analyst in a financial role may no longer be responsible for processing numbers, but they still own the thinking – the judgment to know what the numbers mean, the instinct to ask questions models didn’t think to ask, and the credibility to walk a board through uncertain decisions.

This is not a new phenomenon. We’ve watched organizations get this wrong during every major technology cycle of the past three decades. The pattern is the same: change in technology equates to a change in headcount. But it’s time we started asking better questions. If AI absorbs these tasks, what does that liberate our people to do?

Workers are already telling us something important. Our recent survey of 2,000 workers in the US and UK found that nearly half (46%) using AI tools have never received formal training. Of those without guidance, 47% taught themselves through trial and error, 36% deliberately limit their AI use to avoid mistakes, and 17% simply pretend to use it when asked.

When asked which skills will matter most to their careers, workers ranked critical thinking, judgment, creativity, and resilience at the top. Technical AI knowledge came last. It’s clear that these workers already understand something their organizations haven’t operationalized – the durable value in an AI-augmented workplace lies not in the technology itself but in the quality of human decision-making brought to its output.

The advantage of treating AI as a release valve comes from investing deliberately in four interconnected capabilities. None requires a transformation announcement, and all compound over time.

To make the most of this opportunity, organizations should first make their workforce visible to itself by building a real-time picture at the skills level – not job titles but actual capabilities. This surfaces where people are developing, where gaps are forming, and which adjacent capabilities could be activated to meet new needs.

Secondly, they should close the distance between learning and work by embedding development in the work itself. AI agents can surface the right guidance at the exact moment a gap appears, triggered by performance signals rather than calendar cycles.

Thirdly, organizations should redesign roles around what AI cannot do. Before making any workforce decisions, three questions deserve honest answers: Which tasks does AI handle well enough to absorb entirely? Which tasks improve when humans and AI work together? Which tasks become more valuable precisely because AI handles everything around them?

Finally, they should invest in managers as the connective tissue. Technology can surface insights and personalize development, but managers control what work gets assigned, how feedback lands, and when someone is ready for a bigger challenge. Developing managers who recognize capability gaps and coach toward judgment rather than task completion turns them into development multipliers for the entire organization.

Every technology disruption I’ve led through has required the same starting point: get honest about the task, not the job. The answers are almost never “entire job eliminated.” They are almost always “this task absorbed, that task elevated, this new task created.” You cannot lead a transformation you haven’t mapped.