The Dusk of Hard Skills

AI is not making expertise disappear. It is making average execution abundant. That changes who needs deep technical skills, and what everyone else is actually paid to do.

For decades, white-collar work followed a fairly simple bargain: learn a hard skill, become useful, get paid. Write. Code. Analyse. Design. Model. Plan. The skill was the work. That bargain is being renegotiated.

AI can already perform a large share of the competent, repeatable version of these tasks. Not the frontier version. Not the version where one bad decision collapses a bridge, a financial system or, less dramatically, production on a Friday afternoon.

The much larger middle: the decent draft, the routine implementation, the plausible visual, the structured analysis, the comprehensive plan. That middle employs a lot of people.

Hard skills are moving upstream

Hard skills will not disappear. Their distribution will change. A relatively small group will still need extraordinary technical depth. Researchers, infrastructure engineers, scientists, model builders and domain experts working in places where novelty matters and errors are expensive will remain highly valuable. Their skills may become even more important because increasingly large systems will depend on them.

For everyone else, technical execution becomes something closer to infrastructure. Most people use electricity. Very few design the grid. Knowledge work already follow a similar pattern. A small number of people will build and maintain the deepest technical layers. A much larger number will operate those layers through models, agents and increasingly capable software.

The average worker does not become less intelligent. They become less directly responsible for producing every intermediate artifact by hand.

When the baseline became cheap

This is not a scientific timeline. It is a rough map of the moments when, in my own point of view, the competent baseline started to feel economically abundant.

The dates are debatable. The direction is harder to ignore.

“Soft skills” are becoming the control layer

The term soft skills is starting to look slightly ridiculous.

They were considered soft because technical execution was scarce. When execution becomes abundant, deciding what deserves to be executed becomes the difficult part.

• Creativity is not the ability to generate more options. Models are already very good at that. Creativity is noticing the option that changes the frame.
• Taste is not decoration. It is a rejection system.
• Interpersonal intelligence is not simply being pleasant. It is the ability to move people through ambiguity, conflict and uncertainty without turning every disagreement into a six-week governance process.
• Judgment is deciding before every variable is known, then remaining accountable for the result.

These capabilities used to sit around the hard skill. Increasingly, they sit above it.

More work becomes In/Out work

I call it In/Out work, at least for now. Information comes in. It gets interpreted, reframed and sent back out, often with a decision attached. This is already how many managers and executives spend their days: calls, emails, documents, dashboards, updates, approvals and conversations. They may not produce a visible object themselves. Their output is direction, alignment and closure.

AI pushes more jobs towards this pattern. The model writes the document, proposes the campaign, generates the interface or implements the feature. The human receives the result, understands the context around it, decides what matters and sends the next instruction.

The best In/Out work produces coherence. The worst version simply forwards PDFs, PowerPoints, and more and more Markdown files.

The human moat is a combination

No single soft skill will save anyone. “Be creative” is not a career strategy. Neither is “have taste,” despite our collective determination to mention taste in every conversation for some time now.

The defensible combination is more interesting:

• Domain understanding,
• Curiosity,
• Taste,
• Social trust,
• Agency.

The ability to frame problems, direct machines, evaluate output, persuade humans and take responsibility when the answer is unclear.

This is good news for multidisciplinary builders. A person who understands product, design, technology, marketing and human behaviour can connect outputs that remain fragmented when treated separately.

The moat is not knowing a little about everything. It is knowing enough across several domains to see the whole system, while having enough depth somewhere to avoid becoming professionally decorative.

The work moves up a level

Hard skills used to be the main unit of production. They are becoming leverage.

For many knowledge workers, the real job will be choosing what matters, creating direction, evaluating what comes back and getting humans to move together.

Hard skills will survive. The mass requirement to perform all of them manually may not.

The future of work could be extremely technical underneath and strangely human on top.