A course can teach a useful technique and still leave someone poorly prepared for the next change. That happens when the learner remembers the sequence of clicks but cannot explain the problem, judge the result or adapt when the interface behaves differently.

What happened

On 6 September 2026, Daniel Susskind challenged the idea of future-proof skills in an education essay.

For additional context, the OECD’s July 2026 skills report discusses foundational abilities, including literacy and numeracy, alongside digital and social skills in an AI economy. The report provides that broader framing.

These are different contributions: a commentary that prompts a question and research that helps situate it. Neither supplies a guaranteed list of occupations or abilities that will remain unchanged. The management task is to design learning that can respond to uncertainty.

Why it matters

Employers often need practical training quickly. A team has acquired a tool, a process is changing, and people want to know what to do on Monday. Specific instruction is useful in that situation.

The problem begins when short-term proficiency is treated as the entire development plan. An employee may become fast at generating a report without becoming better at deciding what the report should establish. If the output changes unexpectedly, their confidence may exceed their ability to investigate it.

I would build training around a complete task. Ask the learner to define the intended result, identify suitable information, use the approved tool and inspect the output against clear criteria. Then change one condition: an incomplete brief, contradictory figures or a customer request outside the usual pattern.

The purpose is to practise adaptation with support. This is a proposed learning exercise, not an assessment that should secretly determine someone’s job security. People need room to disclose uncertainty and receive useful feedback.

The bigger shift

My view is that the unit of learning should become a repeatable decision process rather than a supposedly permanent tool trick.

For a marketing analyst, that could mean explaining the business question before generating a chart, checking the provenance of the numbers and deciding what conclusion the evidence can support. The software may change. The learner still needs a method for making the work trustworthy.

This approach also creates a maintenance responsibility. Someone must review examples when a tool changes, retire instructions that no longer apply and collect the situations learners find difficult. A training programme without that ownership can become an archive of outdated demonstrations.

Managers should distinguish three kinds of progress: greater confidence using a tool, better work produced with it and stronger independent judgement. They can develop together, but one does not prove the others. A useful review asks for evidence of each.

That is consistent with starting AI strategy from business needs. The training objective should explain which decisions or outputs improve, rather than merely recording attendance.

My take

I would be cautious about any promise to make a career future-proof through a single course. A more credible promise is to help someone practise a useful task and become better at learning the next version of it.

Choose a recurring piece of work and create a short cycle: attempt, inspect, discuss and revise. Include an experienced colleague who can explain why a result is strong or weak.

Then return to the task after a meaningful change in the tool or workflow. The organisation will have something more useful than a fixed list of fashionable skills: a way to keep its capabilities current.