Qatreh AI Logoقطره
Back to blog
AI and the Future of Work

Preparing a Team for Work That Changes

The least-quoted finding in the Anthropic report: what changed is the route into work, not existing jobs. Three skills rise in value, none of them programming.

تیم قطرهSeptember 6, 20265 min read
Preparing a Team for Work That Changes

One finding in Anthropic's recent labour-market research gets quoted less than the rest, and matters more to managers than any of them: what has changed most is the route into work, not existing employment.

The short answer

The report finds no systematic increase in unemployment among highly exposed workers since late 2022, but among workers aged 22 to 25 entering highly exposed occupations the job finding rate fell by roughly 14 percent. The management reading is clear: the simple entry-level tasks new staff used to start on are the same ones that move first — so the way an organisation trains people has to be redesigned.

Why this is an organisational problem, not a jobseeker's

Organisations usually start new staff on simple work: entering data, tidying reports, answering repeat questions. Those tasks did two jobs at once — the work got done, and while doing it the person learned the business.

When that work moves, the work still gets done and the learning stops. That is not an HR problem; it is a question of what capacity the organisation has in two years.

The practical answer is to replace that route with structured training — something that used to happen free and implicitly, and now has to be designed on purpose.

The skills whose value rises

Putting the report's pattern beside project experience, three groups of skill rise in relative value:

  • Defining the problem and the measure. Turning "this is slow" into "this figure must move from x to y". No tool substitutes for it, because it depends on knowing the business rather than the technology.
  • Judgement on edge cases. Once ordinary cases are automated, what is left for a person is precisely the hard ones. An employee who only knew the ordinary cases needs training for the hard ones.
  • Evaluating output. Recognising whether a system's answer is correct or merely reads fluently. This is a new skill and nobody has it naturally.

None of the three is programming. That matters: most organisations begin training with "let's learn Python", while their gap is somewhere else entirely.

What the report says about unexposed work

About 30 percent of workers have zero exposure — cooks, mechanics, bartenders. Work requiring physical presence and manual skill has barely moved.

For an organisation holding both kinds of staff, that means a single uniform training plan is wrong. Teams working with documents and data need training now; teams working with equipment need different training on a longer horizon.

A training plan that matches reality

What works in practice is not a general course. Our suggested order:

  1. Write out the unit's task list and mark each as automatable or augmentable. The method is in automation or augmentation.
  2. For augmentable tasks, train on the tool — against the work the team actually does, not a generic example.
  3. For automatable tasks, train on evaluation — because that person's role shifts from doing to reviewing.
  4. Make one person the owner of the measure. Someone who understands what the system does and where it struggles. Without that person, every small later change becomes a new contract.

The numbers behind this advice — why real usage sits at a third of available capability — are in the 94-to-33 gap.

For organisations in Karaj and Alborz

For many organisations in Karaj and Alborz province the start of the academic year is when the training budget is set. The practical suggestion: rather than booking a general course, produce the task list first and then decide what training is needed.

The difference in outcome is large. A general course has a fixed syllabus identical for everyone; a programme that starts from the organisation's own process applies to real work on day one.

We run that task-list session on site, at Alborz Science and Technology Park or at the organisation's premises, because working out which skill is genuinely missing cannot be done from an organisation chart.

Frequently asked questions

Does AI cause unemployment?

Anthropic's research reports no systematic increase in unemployment among highly exposed workers since late 2022. What did show a change is the job finding rate for workers aged 22 to 25 in highly exposed occupations. Note this data is from the United States labour market.

Does our team need to learn programming?

Usually not. Most organisations' gap is in defining the problem, judging edge cases and evaluating output. These can be taught without programming knowledge.

How long does training take?

It depends on scope, and no figure is meaningful before the task list exists. What is constant is that training must run alongside deployment, not after it.

What if staff are worried about being replaced?

Early transparency. Employees usually know better than managers which part of their work is repetitive. Involving them in the task-identification stage makes the identification more accurate and the worry smaller.

Who should own the training programme?

A manager who also has the authority to change the process. Training designed separately from process change does not survive contact with daily work.

What is the difference between a general course and enterprise training?

A general course has a fixed syllabus identical for everyone. Enterprise training starts from your own process: first it is settled which work is to become intelligent, then the syllabus and exercises are built on your organisation's own data and workflow.

Related services from Qatreh