Vision for GenAI integration into business education

Business students entering the UK labour market face two challenges which previous years’ graduates were spared.

First, the number of positions being advertised has fallen substantially. The drop has been more severe for AI-exposed job roles than for those not exposed to AI*. AI-exposed job roles include those in business, marketing, HR, data & analytics, law, accountancy, finance, research, legal, media, and software development. Employers are actively exploring ways to substitute entry-level human labour with AI.

In other words, business-school graduates are attempting to join the labour force at a time when there are fewer entry-level jobs, thus greater competition, and pressure within firms to reimagine their operations to take advantage of labour-saving AI.

Graduates are hit with the second challenge when they are lucky enough to land a job interview: ‘Tell me what you have done with AI?’ When businesses are surveyed about their use of AI, they lump techniques from several different epochs together, ranging across classical statistical techniques,* more modern Machine Learning classification techniques, * as well as Generative AI (BoE 2025). Most business-school degrees provide exposure, but not deep training in these families of techniques, except for in analytics-, data science-, statistics-, and econometrics-heavy programmes. Quick-witted graduates from economics, finance, operational research and cognate programmes can spin an answer around classification using logistic regression, but this option is not available to most business students. In any case, the ‘AI’ that excites employers is Generative AI, and not so much the application of classical statistical techniques.

Whereas the first two families of techniques will continue to be applied by those with deep specialised expertise to develop AI systems for company-wide deployment, the threshold for being able to build personal productivity-enhancing tools with GenAI is much lower. This is partly because LLMs can be used to produce and test code, but also because such LLM-assisted development is iterative, and has clear success/fail outcomes - either the code compiles and, through testing, can be shown to perform the task or operation it was developed to accomplish, or not. If not, additional cycles of refinement and testing can be undertaken until satisfactory performance is achieved.

For an employee with initiative and willingness to engage with the technology, it is perfectly feasible for them to improve their own on-the-job efficiency by building AI workflows and AI agents to automate specific job tasks and processes. But there is a learning curve, and ideally a graduate should already have Agentic AI Building as part of their skills and experience upon seeking to enter the labour market.

Let us consider two educational models: one in which AI agent building skills are integrated into business-school education, and one in which AI agent building skills are developed in a post-degree ‘finishing school’.

Integrated model 

GenAI poses a fundamental challenge to the efficacy and integrity of the self-directed learning in all forms of take-home assignment, including notably essays and dissertations (Walsh 2025). In 2025 California State University (CSU) and Ohio State University (OSU) became first movers in embedding AI literacy throughout their programmes. The present ‘integrated model’ is more ambitious and transformational than the CSU/OSU GenAI literacy, prompt crafting, and critical judgment approach.

First, reimagine the organising principles that define specialist masters degrees. Instead of being solely composed of learning modules defined by topics and themes developed within a particular academic literature, a fraction of learning modules are defined around specific job roles, job tasks, and job processes.

Within these role-, task-, and process-focussed learning modules, lectures cover not only the workplace context in which the role/task/process is carried out, but also relevant theory and empirical data. The relevant theory will typically not be from one single academic discipline alone. Empirical data will be brought in not only to inform the characterisation of the sectoral and workplace context, but also the job role/task/process itself.

Conventional seminars and tutorials are replaced with development workshops, in which the instructor walks the students through, and student follow along on their own laptops, the building of an AI agent, or an AI workflow to automate part of the job role/task/process that is the subject of the module.

A learning module early in the academic year might only construct a single, simple AI workflow for the pedagogical purpose of familiarisation with the programming environment. The number, complexity, and functionality can be increased as the academic year progresses and students accumulate experience. Later modules will also feature testing procedures as well as improvement iterations of the code to ensure reliable, robust performance.

Dissertation modules can also be organised around the building, testing, and business-use evaluation of AI agents and workflows.  

Finishing-school model 

For recent graduates who have not immediately succeeded in the labour market, a compressed-duration ‘Agentic AI Finishing School’ can boost the attractiveness of a graduate’s value proposition as an employee.

The modules in the finishing school could be more generic than those in the Integrated Model, focussing more on the technical development of the AI Agents and AI Workflows - but not as much as existing online course providers, of which there are many. Each module must focus not just on interesting technical implementations, but on the underlying job role/task/processes being automated, the associated guardrails, internal quality checking loops, and testing procedures to take the technology closer to being implemented in a real business context.

Delivery could be online, or hybrid online and in-person.

References 

[*] This is the case both in the UK, as documented by Allas and Goodman (2025) at McKinsey, as well as in the US, as documented by Simon (2025) at Revelio Labs and Constantz (2025) at Bloomerg.
[*] logistic regression, principal components analysis, k-means cluster analysis
[*] neural networks, support vector machines, decision trees, ensembles of weak classifiers, bagging, boosting, deep learning neural networks

BoE (2024) Artificial intelligence in UK financial services – 2024. Bank of England and Financial Conduct Authority, 21 Nov 2024.

Constantz (2025) Is AI killing entry-level jobs? Here’s what we know. Bloomberg, 30 July 2025.

Allas and Goodman (2025) Not yet productive, already disruptive: AI’s uneven effects on UK jobs and talent. McKinsey UK, 14 July 2025.

Simon (2025) Is AI responsible for the rise in entry-level unemployment? Revelio Labs.

Walsh (2025) Everyone Is Cheating Their Way Through College. ChatGPT has unraveled the entire academic project. New York Magazine, 7 May 2025.