Originally published by Grace Lordan in LSE Business Review on September 14, 2026, under a Creative Commons Attribution 4.0 licence. Adapted here under that licence: rewritten for a general audience, with new structure, new explanations and a new ending. The original Shutterstock image is not reused. LSE Business Review and the author do not endorse TeaLamp. Prepared by the TeaLamp Editorial Team with the help of AI.
Artificial intelligence is changing what we do at work, but Grace Lordan, a professor at the London School of Economics, warns that a quieter change may matter more: a soft skills crisis, in which the human skills we stop practising, like judgment, critical thinking and dealing with disagreement, quietly fade. She argues that we are sleepwalking into it.
Key takeaways on the soft skills crisis
- In a 2026 survey of more than 1,000 workers in the United States, 57 per cent picked AI’s erosion of human skills as the year’s top workforce worry, ahead of jobs being displaced.
- Lordan names four skills at risk: deep analytical writing, spotting a problem before anyone frames it, tolerating intellectual friction, and keeping knowledge in memory rather than looking everything up.
- A Microsoft and Carnegie Mellon study of 319 knowledge workers and 936 real examples of AI use at work found that trust in the AI went with less self-reported critical thinking, while trust in one’s own skill went with more.
- The skills Lordan says matter most now are editorial judgment for checking AI work, relational intelligence for the human moments, and agility and resilience for constant change.
- Her advice to leaders is practical: train people properly in AI, protect the junior work that builds judgment, and measure the human capabilities that no dashboard currently tracks.
The conversation nobody is having
Public discussion of AI at work mostly circles three questions: which roles vanish, which fresh technical skills staff need, and how quickly firms should buy the newest tools. Lordan argues that a different question goes unasked: what becomes of the human abilities we stop exercising once a machine handles that part of the work?
Where firms have acted, the money has mostly gone to tool training: teaching people how to prompt, how to integrate AI and how to automate. What gets little funding is the human side that decides whether the tools produce anything worthwhile: sound judgment, critical thinking, spotting a polished answer that is wrong, and the people skills that keep a team together.
The twist is uncomfortable. The very abilities machines cannot copy are the ones our training budgets are letting rust, while the tools themselves get all the attention.
Four skills quietly slipping away
The first skill at risk is deep analytical writing. Staff who let AI write every first draft miss the mental exercise of building a complicated argument from nothing. As Lordan puts it, writing is thinking you can see; hand the writing over and you hand over a large part of the thinking too. Cited research put the productivity gain for greener workers at up to 34 per cent, while veterans saw little gain and sometimes worse output.
The second is working through uncertainty with no prompt to lean on. AI tools shine once the question is clear. But the skill that pays most in an organisation is noticing a problem before anyone has worded it: sitting with unease, spotting what is absent, naming the true issue. That sense never grows if the first move is always to ask a chatbot.
The third is staying power in hard conversations. Real teamwork needs pushback, bargaining and the slow, sometimes awkward job of fitting together views that do not line up. AI makes agreement feel effortless, and that ease teaches people to dodge the very discomfort where new ideas are born.
The fourth, and for Lordan the most worrying, is memory and context slipping. When any fact can be searched and each choice comes with a machine suggestion, there is less reason to build deep knowledge of a field. Knowledge you carry in your head is different from knowledge you fetch: the first supports pattern-spotting, gut instinct and fast judgment in context, while the second only hands back a ready-made summary someone, or something, else already assembled.
A blank notebook and a pencil: the tools of thinking before the first draft is handed to a machine. Image generated by AI.
What the research shows
The warning is not only an opinion. A study by Microsoft Research and Carnegie Mellon University, presented at the 2025 CHI conference, surveyed 319 knowledge workers about 936 real examples of using generative AI at work. The study’s core result concerned trust: people who trusted the AI more said they thought less hard about its output, while people who trusted their own skill more said they thought harder. It relies on what participants reported about themselves; it does not track skill over years.
The cost of unchecked AI output is also visible in public cases. In 2025, Deloitte in Australia revised a government report after it was found to contain fabricated citations and a quotation attributed to a court judgment that did not support it, and the firm partly refunded its fee. A Deloitte report for a Canadian provincial government was likewise found to contain citations that did not exist. In 2023, two New York lawyers faced sanctions after filing a brief packed with invented case law produced by ChatGPT. In each case the shape repeats: the machine produces something plausible, and a person signs it off.
None of this means AI is useless. Lordan’s own work with Daniel Jolles put the weekly saving at 7.5 hours for AI users. Yet 68 per cent had no training whatsoever, and the trained group produced twice as much. Age explained none of it; spending on training explained all of it.
Why people skills are the safest bet
Her patent-data research points the same direction. Skills built on winding, non-straight-line thought resist automation best. Whole jobs that mix dealing with people and analytical work, negotiation, strategy, leading a team, face the lowest displacement risk. Over time, those team-leading skills grew markedly more prized, while technical skills kept being swapped for newer ones.
Three capabilities to double down on
So what should workers and employers protect? Lordan names three capabilities. The first is editorial judgment: the habit of sizing up AI output, pushing back on it and making it better instead of swallowing it whole. It is not a technical trick. It calls for subject knowledge, a sense of quality and the nerve to call a smooth, confident paragraph wrong when it is.
The second is relational intelligence. As machines take over routine messages, the human exchanges left behind weigh more. A meeting, a hard talk, a real moment of connection all gain weight exactly because they grow scarce. The leaders who do well, she argues, will guard human contact as a rare asset rather than a drag on efficiency to be engineered out.
The third is agility and resilience. A process can be automated, but no machine can judge whether that process deserves to exist next month. Changing course, staying calm amid uncertainty and steadying a team through round after round of change remain deeply human strengths, and they are scarce.
A staircase rising toward the light, like a career built one practised step at a time. Image generated by AI.
Why it matters
The risk is not that AI will replace everyone. It is that, job by job and task by task, we will stop doing the thinking that makes us good at our jobs, and only notice when a difficult case arrives and nobody is practised enough to handle it. Junior employees are especially exposed: entry-level postings in the United States are down more than 35 per cent since January 2023, and the everyday jobs that once showed newcomers how a company works are disappearing. If the lowest rungs of the ladder are automated away, Lordan warns, the pipeline of future leaders will dry up.
Her fix begins with counting what counts. At present nobody measures the human abilities that may be thinning as AI spreads. Without a way to gauge critical thinking, context judgment and joint problem-solving across a workforce, leaders steer into the biggest skills shift in a generation with no instruments. Training matters too, but only training that teaches doubt, not just prompting.
Related reading
For another story about how work is changing, read Global Poverty Is Falling at Every Line, Not Just One, which looks at what long-run economic data can and cannot tell us about progress.
Frequently asked questions
Are soft skills really declining because of AI?
The evidence so far is about reported behaviour, not proven long-term decline. Workers in surveys say they think less critically when they trust AI output, and employers report little training in judging AI work. Lordan’s argument is that the risk is real enough to act on before it is measured in full.
Does this mean companies should slow down AI adoption?
No. Lordan’s point is that adoption without judgment is the problem. Her prescription is to pair AI tools with proper training, protected practice for junior staff, and checks on AI output, rather than to stop using the tools.
Which human skills matter most now?
Editorial judgment for checking and improving AI work; relational intelligence for the human moments that matter more as routine contact is automated; and agility and resilience for changing course and leading through repeated change.
Sources
- We are sleepwalking into a soft skills crisis caused by AI, LSE Business Review, by Grace Lordan, published September 14, 2026.
- LSE Blogs Republishing Policy, confirming LSE Blogs articles are available under a Creative Commons Attribution 4.0 International licence unless otherwise stated.
- 63% of Workers Say AI Will Make the Workplace Feel Less Human in 2026, Resume Now survey of 1,003 employed United States adults, confirming the 57 per cent figure for AI reducing human skills as the biggest workforce issue.
- Using AI, Like ChatGPT, Damages Critical Thinking: Study, on the Microsoft Research and Carnegie Mellon study of 319 knowledge workers and 936 examples of AI use at work.
- AI vendors and consultants: lessons for Canadian employers from Deloitte federal contracts, Canadian HR Reporter, confirming the Australian partial refund and the Canadian report with citations that did not exist.
0 Comments