Our college is building its own AI tools instead of relying on third parties

Every college I speak to is doing something with AI. Far fewer are building anything of their own.

At City of Portsmouth College (CoPC), we have built our own large language model interface, powered directly by API and connected securely to our digital ecosystem behind two factor authentication. It sits alongside a conviction I have held for a while now, that the most valuable AI work in a college is not a general chatbot bolted onto the intranet. It is a set of narrow, purpose-built tools aimed at the processes that are quietly losing you time and accuracy.

In my last role I commissioned and built 20 bespoke agents doing exactly that, each one aimed at a workflow where productivity was slipping or error rates were creeping up. Now I am in post at CoPC, we will be exploring automation and AI solutions for similar workflows over the coming months. It is worth setting out plainly what this actually means, because “AI agents” has become one of those phrases that gets nodded at in meetings without anyone quite agreeing what it refers to.

An AI agent, in the sense I mean it, is not a general-purpose tool like the one staff might use for drafting an email. It is a piece of AI configured to do one job, with access to only the data it needs for that job, sitting inside a workflow that a human still checks. Think less “ask it anything” and more “it handles this one process, and someone reviews what it produces before anything happens as a result.” The agent does not replace the member of staff who owns that process. It removes the repetitive part of their job so they can spend more time on the part that actually needs a person.

The workflows worth targeting are rarely glamorous. They are the ones eating up disproportionate staff time relative to their complexity. Routine data queries, checks that used to mean chasing three different systems, processes where a small error early on causes a much bigger headache further down the line. None of this makes for an exciting demo, and that is rather the point.

The case for building rather than buying comes down to control, cost and trust. Building our own secure AI platform, rather than relying on a third-party interface, means we pay only for what we actually use. We decide exactly what data the model can see, and we design human checkpoints into every workflow ourselves.

Sitting behind our own authentication means we are not trusting someone else’s assumptions about data governance, and we are not locked into whatever roadmap a vendor chooses next. For a sector already nervous about GDPR and the Data (Use and Access) Act 2025 – not to mention intellectual property (IP), that combination of being affordable, safe and secure matters more than any feature list.

None of this is straightforward though. Building agents well requires technical capability that most colleges do not currently have, so either recruiting someone or finding a commercial partner is essential and a cost.

There is a genuine risk of a widening gap between colleges with the digital maturity to do this and those without it, which is exactly why I advocate for sharing experience, knowledge and resources rather than every college reinventing the wheel alone.

Staff also need to trust these tools, which means involving them in the design rather than presenting them with a finished product. An agent built without staff input, however well engineered, tends to get quietly ignored.

My honest advice to any leadership team considering this path is to resist the urge to start with the technology. Start with the process that frustrates your staff the most, the problem that you are trying to solve, understand it properly, and only then ask whether an agent could help.

Building bespoke AI solutions is not about chasing a trend. It is about deciding that your organisation’s understanding of its own problems is worth more than a vendor’s generic solution to a problem you have not actually defined yet.