An entry-level role is both a job and a route into more difficult work. If a company changes the tasks juniors perform, it should also examine how they will develop the judgement expected of experienced colleagues.

What happened

On 12 September 2026, Guardian analysis described weaker programming employment outcomes for UK computing graduates.

The underlying official data needs its own time context. HESA’s latest Graduate Outcomes release was published on 4 June 2026 and covered the graduating class of 2023/24. It is therefore historical evidence about graduate outcomes, not a live count of vacancies in September. HESA provides the release details.

A change in outcomes does not, by itself, establish how much AI caused it. The practical question for employers is what kind of entry pathway they are building as their own work changes.

Why it matters

Suppose an engineering team uses AI to prepare routine code changes. That may alter the assignments previously given to a newcomer. The team still needs to decide how that person will learn to understand an unfamiliar system, diagnose an error and explain a proposed change.

Leaving learning to chance is a weak response. So is preserving every old task regardless of whether it remains useful. A more deliberate approach identifies the capability each assignment was meant to develop and designs a suitable replacement when necessary.

For example, a junior might inspect a generated change, explain its assumptions, create a test for an overlooked condition and discuss the result with a reviewer. This is a proposed exercise, not evidence that any particular employer uses it successfully.

The reviewer needs time to teach. If the organisation removes routine work while also removing access to experienced colleagues, it may create a gap between the responsibility it wants juniors to take and the practice available to them.

The bigger shift

My view is that workforce planning should include the production of expertise. A company can optimise its current output while weakening the route through which it develops future senior staff.

That risk is worth examining rather than assuming. Track what newcomers actually do during a normal week. How much of their work exposes them to customer needs, system behaviour and meaningful feedback? How often do they explain a decision instead of merely submitting a finished artefact?

A useful development plan would include supported practice, progressively harder assignments and clear evidence for increasing responsibility. AI assistance can be part of that plan, provided managers assess the person’s understanding as well as the quality of the assisted output.

Hiring criteria may need attention too. If the workplace uses approved AI tools, an assessment can include both assisted work and an explanation of the candidate’s reasoning. The aim is to understand capability under relevant conditions, with the rules made clear in advance.

These choices belong in a business-led AI strategy, because they affect the organisation’s future capacity as well as its immediate efficiency.

My take

Employers cannot resolve uncertainty about an entire labour market through one programme. They can make their own entry routes more credible.

Review one junior role with the people who supervise it. Identify which learning opportunities have changed, which abilities remain essential and where a newcomer can practise them safely.

Then give someone responsibility for maintaining that pathway. A graduate should be able to see how useful work leads to greater capability, and a manager should be able to explain the evidence for that progression. That is a more actionable response than treating every employment headline as a forecast of a whole career.