Stone Centre Workshop on AI, Analytical Disciplines, and Graduate Careers: recap

On 3 July 2026 the Stone Centre at UCL, in partnership with the UCL Policy Lab, held the Workshop on AI, Analytical Disciplines, and Graduate Careers: Opportunity, Risk, and Institutional Responsibility.

‍

The day brought together employers, professional bodies, regulators, and educators across two panels and an afternoon hackathon. The shared question: how is artificial intelligence reshaping what we teach, how we teach and assess it, and what does this mean for graduates entering analytical careers?

‍

To set the scene, the co-organisers Professor Elinor Jones and Professor Parama Chaudhury had asked panellists to reflect on the latest IFS evidence. Graduates still earn more than non-graduates, and analytical degrees remain a strong investment. Yet the reassurance is uneven: around a fifth of domestic graduates might have been financially better off not attending university, and even within the high-paying analytical disciplines, there is a lot of variation in wages. Structural change on the scale of AI could increase that proportion. With a majority of law firms now reported to use AI, the message was that institutions cannot wait for regulation to catch up. The task is to make graduates genuinely AI-literate while maintaining their critical thinking skills.

‍

Panel 1: How Graduate Labour Markets in Analytical Fields are Changing

‍

The first panel was chaired by Professor Cloda Jenkins (Imperial College London) and featured Charlie Ball (Jisc), Elanor Currin (PwC), and Gemma Gathercole (ACCA). The panel cautioned that attributing labour-market shifts to AI is harder than the headlines suggest. Charlie Ball noted that recruitment actually rose in law and accountancy last year even as AI raised productivity, while management consultancy, financial analysis, and the visual arts saw sharp falls. Isolating AI's effect from Brexit, the pandemic, the global financial crisis, or current geopolitical tensions is difficult, and AI can be a convenient narrative for headcount decisions firms would have taken anyway.

‍

A recurring structural concern was the shift from a 'triangle' to a 'diamond'-shaped workforce. As firms trim entry-level roles most exposed to automation, they also shrink the pool from which mid-level talent is later promoted. Gemma Gathercole framed this as a prisoner's dilemma: cutting junior hiring pays off for one firm only if rivals hold firm, echoing how 1990s recruitment cuts produced a shortage of experienced staff years later. Against a backdrop of investment in people at its lowest since 2007, and with a flood of near-identical, AI-polished applications pushing up the cost of hiring on both sides, the panel argued that recruitment is now daunting for applicants and employers alike.

‍

For Elanor Currin, the more accurate frame is augmentation rather than replacement, with entry-level roles being 'seniorised' and outcomes depending on how organisations choose to respond. Human contributions like integrity, judgement, and the trust built between employer and employee become more valuable, and agility and lifelong learning move to the centre of a graduate's working life.

‍

The inequality aspect

‍

The IFS picture points to large distributional effects, with a squeeze on the middle even as the top and bottom hold up. Panellists expected AI, like other major technological disruptions such as the ICT revolution, to widen existing gaps: benefits accrue disproportionately to the already affluent and well-qualified, while digital poverty (unequal access to devices, workspace, and reliable broadband) holds others back. Graduates with disabilities have seen the largest recent drop in recruitment outcomes, likely tied to more risk-averse hiring rather than to AI, and equity and inclusion in both recruitment and assessment surfaced repeatedly. Several speakers agreed AI could narrow some gaps, but only if students are actively taught to use it well.

‍

Panel 2: Institutional and Disciplinary Responses

‍

Joshua Fleming (Office for Students), Professor Rachel Hilliam OBE (The Open University and Royal Statistical Society), Robyn Henriegel (Institute of Physics), and Peter Watkins (CFA Institute) discussed how universities and professional bodies might respond.

‍

Peter Watkins described a shift from information-gathering to judgement, producing a 'hybrid professional' who pairs domain knowledge with enough AI literacy to recognise opportunities and risks, alongside the human skills that hiring managers now rank above all else and most often find lacking in recent graduates.

‍

Rachel Hilliam offered a statistician's reframing. Large language models are, in essence, complex statistical models, so the critical faculties economists and statisticians already teach such as interrogating data quality, bias, and underlying assumptions are precisely what students need. She argued these skills should be embedded at the level of whole qualifications rather than single modules.

‍

Robyn Henriegel argued that the integrity of assessment, and the trust that accreditation protects, should not send universities retreating to in-person exams. The better response is authentic assessment that also captures communication and collaboration, none of which is specific to a degree subject or even to AI. Joshua Fleming, speaking for the regulator, reminded the room that these responsibilities are not new and framed the moment as a chance to correct old failings while holding the line on standards and guarding against grade inflation.

‍

A shared conclusion was that the hard/soft-skills divide is a false dichotomy. Good communication, as Rachel Hilliam put it, is about conveying what actually matters and making technical work interpretable for a client, a colleague, or a generative model. Skills that are hard to assess, the panel agreed, must still be assessed; drop them and students infer they do not count.

‍

The hackathon

‍

The afternoon hackathon organised by Professor Susan Smith (UCL) and facilitated by a team of academic and professional services colleagues turned these themes into curriculum ideas. One cross disciplinary groups of academics, students and professionals worked together to prototype ideas to build “T-shaped” skills; deep disciplinary mastery paired with broad collaboration, delivered through a credit-bearing, compulsory cross-faculty model. Their staged year one to three journey moved from peer-learning foundations, through community and Student Union projects, to global challenges of rising complexity, supported by graduate mentors and a 'Joker' system that lets students swap credits toward weaker areas. It was a concrete attempt to build the adaptable, collaborative graduates the panels had called for, and a reminder that the institutional response to AI is, ultimately, a question of design. Several outcomes are planned to socialise these outputs from the hackathon.

‍

We are thankful to all panel members, attendees, and the UCL Policy Lab for making this event possible.

‍

To read more on this subject, see our recap of last year’s Stone Centre Workshop on AI and Economics and the associated RES report.