Fifty-seven per cent of work hours are automatable with today’s technology. McKinsey published that figure, and it landed in every boardroom presentation, every all-hands slide deck, every nervous conversation at the coffee machine. It is, by any reasonable reading, a terrifying number — and it is also, by any reasonable reading, almost entirely beside the point.
Because what actually happens when you automate 57% of someone’s work hours isn’t that you eliminate 57% of the people. You change what all of them do. Seven in ten professional skills are evolving right now, not vanishing, and the distinction between those two things is the difference between an organisation that panics and one that prepares. Most are panicking.
What endures, and why it matters more than you think
The skills that survive AI aren’t the ones you’d put on a competency framework. Persuading a sceptical stakeholder, reading the tension in a room before it becomes a conflict, holding a difficult conversation with someone whose role is about to change — these are irreducibly human capabilities, and they’re growing in demand precisely because the routine work around them is disappearing. When the administrative burden lifts, what’s left is the relational complexity that no model can navigate.
Registered nurses are the clearest example, though every profession has its version. The documentation, the scheduling, the routine monitoring — all of that is automatable. The judgment call at 3am when a patient’s vitals don’t quite look right but don’t quite trigger an alert? That stays. And it becomes a larger proportion of the job, which means the skill required to do it well becomes more, not less, valuable.
Your organisation’s responsibility here is uncomfortably specific: you need to be able to tell your people which of their capabilities fall into this category. Not in the abstract. For their role, in their team, this quarter.
What transforms — and the trap most organisations fall into
Writing doesn’t disappear; it becomes a different kind of skill. The mechanical act of drafting — first pass, blank page to structured document — gets absorbed by AI almost entirely. What remains is knowing what you want to say in the first place, shaping argument and tone with precision, and catching the subtle thing the model missed because it doesn’t understand your audience the way you do. The Anthropic research on this is unambiguous: 80% of task time vanishes when you redesign work around AI assistance. But that number has an enormous asterisk attached to it, because it only materialises if someone actually rethinks the process. Effort and enthusiasm alone won’t get you there.
Coding follows the same pattern — architecture and debugging become the skilled work; syntax and scaffolding get automated. So does financial analysis, so does market research, so does half of what your legal team does on a Tuesday afternoon. In every case, the mechanical layer compresses and the judgment layer expands. The trap is assuming this happens automatically. It doesn’t. It happens when someone redesigns the workflow, redefines the role expectations, and gives people permission to work differently. Without that deliberate intervention, you get the worst of both worlds: AI tools sitting unused on screens while people carry on doing things the old way, slightly guiltily.
The bottleneck nobody budgeted for
Here’s what both major studies converge on, and what almost nobody has planned for: the time savings from AI don’t automatically become productivity gains. They become supervision requirements.
When agents handle task execution, humans handle everything that sits around it — deciding which tasks agents should touch in the first place, checking that outputs match requirements, catching edge cases, making scaling decisions, and maintaining the quality standards that your clients expect but your AI doesn’t inherently understand. Supervision, in other words, becomes the work. And most organisations haven’t designed roles, team structures, or management layers for a world where that’s true.
The structural implication is counterintuitive. You might need a flatter but wider organisation — more people overseeing smaller, more specialised pieces rather than fewer people trying to manage everything. That’s a fundamentally different management architecture from the one most companies operate, and nobody’s going to stumble into it by accident.
The equity question that leadership keeps avoiding
High-wage workers automate first. That fact contradicts almost every narrative your workforce has absorbed about AI — the assumption that automation hits the factory floor and the call centre before it reaches the corner office. It doesn’t. Accountants, software developers, legal professionals: these are precisely the roles where AI agents have the deepest impact, because these roles involve the structured, knowledge-intensive work that models handle best.
Which raises a question that most leadership teams are not yet willing to answer honestly. If a senior role becomes 30% smaller because you’ve redesigned the process around AI, what happens to that person? Their trajectory, their compensation, their sense of professional identity? These aren’t forced outcomes — organisations have genuine choices here — but they require deliberate decision-making at a level of specificity that most boards haven’t reached. And every week you delay that conversation, your most capable people are drawing their own conclusions in the absence of yours.
What your people are actually worried about
The anxiety in your organisation isn’t really about AI. It’s about clarity — or, more precisely, the absence of it. Where are capabilities heading? What does the organisation expect from people between now and that future state? What should someone be learning right now to remain valuable in eighteen months? These are questions your workforce is asking whether you answer them or not; the only variable is whether they get good answers from leadership or anxious ones from each other.
The data says skills will evolve. Your job — and I mean this as specifically as it sounds — is naming which ones, in your context, are evolving toward what, and what your people need to build right now. Not a generic reskilling programme. Not a learning platform with 400 courses nobody completes. A conversation, team by team, about what’s changing and what’s expected.
Ask them what skills they think are shifting in their work. Then listen — not for the polished answer, but for what they’re actually afraid of. In twenty years of doing this work, I’ve found it’s rarely obsolescence. It’s being left behind while everyone else figures it out.
Signal42 | Beyond the hype. Into the value. |