The Skills That Stay Valuable When Machines Master the Work

In a field experiment with Boston Consulting Group, 758 consultants were given realistic work tasks, and some were allowed to use AI. The researchers then added a task designed on purpose to lie outside what the tool does well. The surprise was that consultants who used it were about 19 percentage points less likely to reach a correct solution than those who worked without it (Dell'Acqua and others, 2023). The tool had not changed; the task had moved beyond the edge of its ability.

A Jagged Frontier

Within the frontier, AI users completed 12.2% more tasks than the others, 25.1% faster, and at higher quality. The researchers called this the "jagged technological frontier": an uneven line separating what the tool does well from what it does not, which the user may not discern, so that two tasks that look alike may fall on opposite sides. The reason is largely that language models produce the probable answer, not the verified one (see the article "How Do Large Language Models (LLMs) Like ChatGPT Work?"). So the first skill is not mastery of a particular tool but knowing where it is right and where it errs.

What Changes in What Is Asked?

Based on a survey of employers, the World Economic Forum estimates that about 39% of current skills will be transformed or become outdated between 2025 and 2030. Analytical thinking comes first among core skills, seen as essential by seven in ten employers, followed by resilience and adaptability, then leadership and social influence. The fastest-growing skills are AI and big data, networks and cybersecurity, and technological literacy. Some 63% of employers said that skill gaps are a major barrier to transforming their businesses, and the Forum estimated that 59 of every 100 workers will need training by 2030.

In the same vein, PwC found in its 2026 barometer that skills required in the jobs most exposed to AI are changing at more than twice the pace of those in the least exposed. That is a measurement of what has happened, not a forecast of what will. In the Kingdom, the Ministry of Education announced in 2025 that the Samai initiative, launched by the Saudi Data and Artificial Intelligence Authority with the participation of the ministries of Education and of Human Resources and Social Development, had trained and certified more than one million citizens, 70% of them in employment.

Who Benefits Most?

Brynjolfsson and his colleagues studied data on 5,179 customer-support agents who were given a generative assistant, and their productivity rose by about 14% on average. The least experienced benefited most in speed and quality, while the most experienced gained little and their quality dipped slightly. The study suggests the tool helps with learning. But it covers a single occupation, customer support, and cannot be generalised to all work.

Habits, Not Recipes

What follows is an editorial distillation of the studies, not a finding stated in them:

• Test the tool on your own work before trusting it, and note where it was right and where it erred.
• Verify numbers, names and references before relying on them.
• Frame the problem before asking for a solution: what is required, for whom, and by what standard will we accept the answer?
• Build depth in one field together with breadth that lets you judge what the tool produces.
• Keep learning, for a technical skill may become outdated faster than good judgement.

Responsibility for a decision stays with whoever signed it; one cannot plead that the machine said so. Trust in God does not negate taking the means, and among them is learning what we need in order to discharge our trust in our work well. Education has a fundamental role in this transition, which we discussed in the article "The End of the Age of Rote Instruction?".