Milburn puts Colleges at the Centre of Youth Transition. Who Prepares their Staff for AI?

Alan Milburn wants colleges to carry young people into skilled work. His final report on young people who are not in education, employment or training (NEET), published on 17 November, makes further education “the principal technical and transition institution” (para 501). But the jobs those learners are heading for are changing with AI. The report sets out workforce expansion and professional development, but does not specify how FE staff will be prepared to teach and assess AI-related judgement or interpret learner data.

What the report proposes for colleges

Colleges would be the principal providers of a new pre-apprenticeship route for 16 to 18 year olds, starting with 25,000 places (para 463). The report aims to double the number of young people achieving an apprenticeship over the next decade. Under-25 starts would rise from around 172,000 to around 270,000 a year by 2036/37, and the achievement rate from around 65% to 80% (paras 461, 462 and 502).

Apprenticeship and FE expansion are to have “first call” on substantial new investment (para 505). As FE expands, college funding would depend more on outcomes achieved than on places provided (para 501).

AI runs through the report. In its section on connecting young people to the jobs of the future, it proposes that training in AI skills be available to every young person, building on the AI Skills Hub. It also proposes a national digital Open College in which AI helps learners find their way through accredited modules (para 458). These are recommendations to government, which has endorsed the report’s direction and accepted some of its central proposals but will not publish a full response until the new year.

The staff gap

The report takes the FE workforce seriously. It proposes a high-status route for experienced professionals to move from industry into teaching, or to combine the two. That route would come with funded, structured initial preparation, high-quality mentoring, subject-specific pedagogy and sustained professional development (para 457). People who know the trade bring what learners most need to see.

What the report does not specify is what that preparation should cover now that AI is changing the trades themselves. Two needs stand out. Staff will have to teach learners to check, correct and take responsibility for AI-assisted work in their trade. And if funding follows outcomes, staff will have to read progression and destination data with care: a learner flagged as at risk of dropping out needs someone who understands what the flag is based on, what it may miss, and what to do next.

Colleges are not starting from nothing. Jisc and the Association of Colleges have published principles for the use of AI in FE colleges, and Jisc offers a strategic framework for college leaders and AI literacy training for staff. That is sector guidance, taken up college by college. What the Milburn report does not do is build AI and data capability into the workforce plan that will carry its reforms.

Industry expertise is necessary, but it is not sufficient. A professional who learned the job before AI took on its routine parts may need help to see how a newcomer learns it now.

How AI is changing early-career work

AI is changing the work through which young people traditionally gain their first experience. In AI and Early Careers (April 2026), Prospects Luminate and the Institute of Student Employers asked 30 employers what they expect. None expected entry-level tasks to be fully automated within three years. Routine content, translation and information-gathering tasks were seen as most exposed, while tasks involving judgement and complex decisions were expected to stay mostly human-led.

Routine tasks can give juniors the chance to develop judgement; whether AI changes that route is an important question for vocational education. David Autor and colleagues gave 133 patent lawyers at 11 US firms an AI drafting assistant for three months (NBER working paper; the trial was registered in advance and run by Google). With AI, the lawyers’ drafts improved on average, and junior lawyers gained most. On a later task done without AI, only the senior lawyers had become better. Junior lawyers showed no average gain.

The authors read this as a sign that existing expertise may be needed to turn AI-assisted practice into lasting skill. That is an interpretation, but it is the right question for colleges to ask.

What the studies show, and what they do not

In 12 German vocational schools, Hanna Brosch and colleagues randomly gave 673 final-year IT apprentices either ChatGPT or no internet access (working paper; the trial was registered in advance). On university-level tasks beyond their training, apprentices with ChatGPT scored 26 to 31 percentage points higher than those without it, with no measured loss of immediate understanding. The study measured performance with the tool. It could not show whether the apprentices learned anything that lasts.

At the University of Sydney, Ali Shakiba and Thomas Chaffey analysed the AI journals of 145 first-year software engineering students (preprint). Of 172 occasions when students stepped in to change what the AI produced, about three in four concerned output that did not suit the task, and about one in six concerned output that was wrong. Judging fitness for purpose is a real skill. Catching and explaining errors is the other half of checking AI output, and the study shows how students intervened, not how many errors they missed.

These studies are suggestive rather than definitive for English FE. Some are working papers or preprints, and most involve learners or workers outside the sector. They raise a question colleges can test: can learners explain, diagnose and correct AI errors, and still perform when the tool is removed?

What the English FE evidence shows

English FE research already documents uneven AI policy implementation and barriers to staff development. Khatibi, Scott and Khandani (2026), writing in Research in Post-Compulsory Education, analysed national and sector documents and the published strategies, policies and inspection reports of six English FE colleges. They found AI policy enacted unevenly, barriers to professional development, concerns about assessment and inequalities in colleges’ capacity to respond.

What remains much less established is which teaching and assessment approaches improve durable vocational capability, and there is little evidence on AI with young people who are NEET. Colleges hold the learners, the employer links and the outcome data that such research needs.

Three things colleges could do this year

Pilot one vocational assessment in which learners critique an AI-generated artefact from their trade, such as a risk assessment, a care plan or a piece of code, explain their corrections, and then repeat a similar task unaided a few weeks later.

Evaluate staff training in AI and data with a scenario-based task, such as reviewing an AI-assisted assignment or interpreting a learner’s risk flag, rather than with a confidence survey.

Partner with a researcher to measure learner outcomes, including whether gains differ between groups of learners, so that the expansion planned between now and 2036 rests on evidence from English colleges.

Government should fund, and evaluate, a professional development programme on AI and data for FE staff, alongside the industry-to-FE route. The report asks colleges to carry the move from school to work for a generation of young people. Their staff need to be ready for the work that move now leads to.

By Prof Rose Luckin, UCL and Educate Ventures Research Ltd

Original source Milburn puts Colleges at the Centre of Youth Transition. Who Prepares their Staff for AI?

Back to home