When the Culture Eats AI for Breakfast

A few months ago, a CEO I have known for years called me, not for a keynote, but to think out loud. He had just let go of nearly a fifth of his workforce. I asked him why. His answer has stayed with me since: “They were AI handicapped.” His company had rolled out licenses, run the training sessions, brought in vendors, ticked every box on the transformation checklist. And yet, when he looked for evidence that work had actually changed, all he found were a few polished slide decks made with AI. The tool had been used to describe the work, not to do it.

I have sat in enough boardrooms and back-of-the-room conversations after keynotes to know this is not one CEO’s problem. It is the story of this moment. I want to use this piece to unpack why, using what I have observed on the ground, and what the research is now confirming at scale.

The Quiet Redesign of Work

Walk into any modern office today and you will notice AI has already slipped into the ordinary software people use every day. Meetings get transcribed and summarised. Invoices get processed without a human first pass. Scheduling, once a small daily tax on everyone’s calendar, now runs itself. None of this looks dramatic. It rarely makes the town hall slide. But it is quietly moving people from creating output to curating it.

I remember running a session for a mid-sized bank in Mumbai where a young analyst told me, half-joking, “I used to write the report. Now I edit what the machine writes, and somehow that feels like less of a job, even though it takes the same three hours.” That sentence has sat with me for a long time, because it captures something research is now putting numbers to. A widely discussed Harvard Business Review piece this year, co-authored by researchers at BetterUp Labs and Stanford’s Social Media Lab, gave this phenomenon a name: “workslop” — AI-generated work that looks finished but does not actually move a task forward, quietly shifting the burden of real thinking onto whoever has to check it. When curation replaces creation without anyone redesigning the job around that shift, people feel busier and less useful at the same time.

Firing the Most Informed Person in the Room

Here is the deeper disruption I keep returning to on stage. For most of corporate history, information was scarce, and scarcity gave certain people power. The one who had read the annual report closely, who remembered the client’s history, who could recall the clause in the contract, was valuable simply for holding what others did not. AI has ended that.

Knowledge is now abundant, instant, and nearly free, which means the org chart built around “who knows the most” is quietly becoming obsolete.

I recall an operations head at a manufacturing firm telling me his best planner, a man who had spent eleven years memorising supplier lead times and machine quirks, went from being irreplaceable to interchangeable within a year of an AI planning tool going live. What that planner had was recall. What the company still needed from him afterward was judgment: the ability to make the final call, read a client’s hesitation, weigh a trade-off the model could not see. Harvard Business School researchers writing in HBR this year make a related point about organisational barriers to AI value capture, noting that most companies do not fail to gain from AI because the technology falls short, but because entrenched roles, incentives, and workflows quietly resist realigning around it. The people, not the software, decide whether the disruption becomes a redesign or a redundancy list.

The Uncomfortable Numbers

I would be doing this topic a disservice if I only told the encouraging parts. The disruption has a harder edge, and it shows up first at the entry level. Harvard economists Seyed Hosseini and Guy Lichtinger, studying resume and job posting data across 66 million workers and more than 280,000 US firms, found that at companies adopting generative AI, junior hiring has fallen by roughly 80 percent per quarter since 2023, even as senior hiring stayed largely untouched. The story is not layoffs; it is quieter than that. Roles simply stop being posted.

Not every voice agrees on the cause. Some economists point instead to interest rate cycles and a post-pandemic hiring correction, arguing AI has become a convenient scapegoat for a slowdown that has other roots, and a few employers, IBM among them, have announced plans to expand entry-level hiring even while deploying AI heavily. The honest answer is that both forces are probably true at once, and any leader who picks only the explanation that lets them off the hook is not really looking.

What is harder to dispute is the productivity picture inside companies that have already adopted AI widely.

An MIT Media Lab report this year found that 95 percent of organisations saw no measurable return on their generative AI investment despite heavy spending, and a separate Goldman Sachs analysis found no clear economy-wide link between AI adoption and productivity gains, even as most large companies now discuss AI on every earnings call.

Automated hiring systems trained on historical data have also been shown to inherit old biases rather than remove them. None of this is an argument against AI. It is an argument against the assumption that installing the tool is the same as doing the work.

The Real Gap Is Not Knowledge, It Is Habit

This is the part I try hardest to get leaders to sit with. In thirty years of watching organisations try to change, I have never seen a workforce this educated, this trained, and still this stuck. The barrier is rarely a lack of understanding. Harvard Business Review’s recent reporting on why AI adoption stalls points to something closer to identity anxiety: employees quietly protecting a sense of relevance and expertise that AI seems to threaten, which no slide about “efficiency gains” will resolve. A separate HBR piece this year found that middle managers, not the front line or the leadership team, are absorbing most of the strain of AI adoption, validating machine output and coaching their teams’ new skills while their own workload and support stay unchanged. That is a textbook recipe for quiet resentment dressed up as compliance.

I described this to a CHRO once as teaching an old dog new tricks, and she corrected me gently: “The dog is fine, Paul. The dog is exhausted from three years of new tricks with no time to practise the last one.” She was right. Change fatigue is real, legacy systems and messy data compound it, and a growth mindset cannot be issued by memo. It has to be modelled by the people at the top, in public, including their own mistakes with the tool.

Peter Drucker’s old observation that culture eats strategy for breakfast has never felt more literal than it does with AI. You can have the clearest AI strategy in your sector and watch it die quietly in a culture that has not been given psychological safety to experiment, fail visibly, and try again. The skill gap headlines get the attention. The habituation gap is where transformations actually go to die.

What I Tell Every Room I Stand In Front Of Now

Three insights keep repeating across every industry I visit. First, automation without redesign only relocates effort; if curation replaces creation, the job description and the metrics attached to it need to change with it, or people will quietly disengage. Second, the scarce resource has shifted from information to judgment, empathy, and context, so promotion criteria built around who “knows the most” need rethinking before the organisation discovers the hard way that they are obsolete. Third, adoption fails for psychological reasons long before it fails for technical ones, so leaders earn the right to ask for change only after they have built the safety to make that change survivable.

The CEO who fired his “AI handicapped” staff was not wrong that something had failed. He was wrong about what. His people were not incapable of learning the tool. They were never given a culture that made using it, stumbling with it, and improving with it feel safe. That is not a training problem. That is not even, in the end, an AI problem. It is the oldest problem in business, wearing a new name.