Backing the next generation: how BCN is helping undergraduates become AI engineers

Most conversations about AI engineering happen between people who were already engineers. This one doesn’t. Billy came to BCN from sales — calls, pipeline, objection handling — and is heading into the second year of his Business Economics degree at Cardiff alongside the work. What follows is a candid account of how far someone can go in a single summer when a business decides to invest in them.

Take me through how you got here, because it isn’t the usual route.

I left school and went straight into B2B sales. A year of that, then Cardiff for a Business Economics degree. One year of the degree down, I took the internship with the Client Innovation Team at BCN. I had no engineering background at all — and BCN took me on anyway, because they saw something they could develop rather than a CV that already ticked every box.

What’s meant the most is what happened after the internship. Rather than it ending when term started, BCN made it clear the door stays open — I can keep contributing and keep learning around university life, and come back in the breaks. For a student, that continuity is rare, and it changes how you think about your own future. It also matters because of how fast AI is moving. If I stepped away for a year I wouldn’t be picking up where I left off, I’d be relearning half of it — so staying close to BCN through uni keeps me current, and BCN gets someone who’s still sharp when they come back rather than starting from scratch.

And the brief when you arrived?

Go out, do what you can, ask a lot of questions, and learn. Training courses if I wanted them. The team right there to use.

Which sounds vague, but it’s actually the most generous thing you can be handed, because it says: we trust you. And crucially, the trust came with support. I asked questions constantly — about things I probably should have known and plenty I had no business knowing yet. Nobody once made me feel stupid for it. That openness is the culture here, and it’s the only reason the rest of this happened.

What has the development actually looked like? You’ve moved fast.

Faster than I’d have believed. I’ve been here three and a half months — one summer. In the first fortnight I had to learn what a branch was. Then what a commit was, and why mine were terrible. By the end of the summer I was the primary builder on a broker origination app for an asset finance client, leading content production on a twelve-month client adoption programme, and building agentic workflows that clients actually run.
That progression didn’t happen by accident. It happened because the Innovation Team gives you real responsibility inside a structured programme, with people around you who want you to succeed. Being trusted with work that matters, that early, has made a huge difference to me personally — not just in skills, but in confidence. I walk into problems differently now.

What taught you the most?

Honestly, not the code. The build for an asset finance client — the broker origination app — and specifically getting the tooling running on a locked-down corporate Windows machine. None of it worked at first. PATH restrictions, npm policy blocks, AppLocker deciding it did not care for my ambitions.

Nobody hands you a runbook for that. You read the error, form a theory, test it, you’re wrong, you go again. Which, it turns out, is the whole job. The pipeline now runs locally using Claude Code, triggered directly from GitHub, so I can set everything up and manage it myself. Small thing. Felt enormous.

You’ve worked across a lot of the business for someone so early in their career.

That’s been one of the best parts, honestly. The Innovation Team sits in the middle of everything, so I’ve worked with Sales on discovery for agentic sales tooling, with the technical teams across Azure, Data and Development, and directly with clients on deliverables. Coming from sales myself, seeing how the technical side of BCN operates — and being welcomed into it rather than kept at the edges — has been an education in itself.
And it goes right to the top. I’ve had access to Execs that I genuinely didn’t expect as a student intern. They’re open, they take an interest, and they encourage you. When senior people give you their time and take your ideas seriously, you raise your own bar. That style — open door, straight-talking, encouraging — runs through the whole business.

What else has landed on your desk?

Client deliverables built straight out of raw material — taking a strategy session transcript and a branded reference file and turning it into an HTML report and a slide deck. I’ve sat in on discovery for agentic sales tooling and consolidated two separate product catalogues into something a sales cohort could actually use.
On the twelve-month client adoption programme, month six alone was three separate content drops: a written knowledge piece, an interactive product demo, and a scenario-based quiz, all inside hard constraints on tone, terminology and demo-data safety. That’s a lot of rope for someone at my stage — but the constraints are the useful part. Constraints are where you find out whether you actually understand the thing you’re building.

And there’s another one since — you built yourself a way to watch your own AI agents.

I run a lot of Claude Code sessions at once and kept losing track of which ones were waiting on me. So I built something that watches them: a small character on my desktop (named after my boss), a green circle for how many agents are working, an amber one for how many need me, and it tells me which repo when one gets stuck. A few days’ work. Saves me more than that every week.
The useful part was a bug I got wrong twice. The third time I stopped reading the code and measured what was actually on screen — and the measurement said the layout was fine, which turned out to be the answer. I’d been looking in the wrong place all along. That habit came from the client work at BCN.

Which brings us to the revision tool. That started as a personal fix and BCN encouraged you to take it further.

I had to retake an exam over the summer — I’d missed the original sitting because I was in hospital, so I was revising around a working week rather than a student one. That’s not the problem revision tools are built for. So I used everything the job had taught me and built one that was.
What I didn’t expect was how BCN responded. Rather than treating a side project as a distraction, the team encouraged it — the entrepreneurial instinct is seen as part of the job here, not competition with it. The intention now is to sell it into universities, and everything about how it’s built assumes someone else’s syllabus, someone else’s marking criteria, someone else’s exam date.

What does it do differently?

Most revision tools give you material. Mine decides what you should revise next, so you stop spending the first twenty minutes of every session guessing. It starts with a placement — two questions per exam section — then builds a plan ordered against the time actually left before the exam. A queue that reorders itself as you get better or as the clock runs down, with a Readiness score per topic derived from how you’ve genuinely scored.

There’s a tutor in it that works through topics by asking questions rather than lecturing, and it never hands over the model answer — the moment a tool does that, you’ve stopped revising and started reading. You can write real exam answers and have them marked against the actual criteria, including photographed handwritten work, because nobody sits an exam in a text box.

Fair question, though: how much of this is you and how much is the model?

The answer is that you find out fast when you’re the only person who can fix it. Marking a photograph of someone’s handwriting against a rubric is not a happy-path problem. Nobody debugs that for you.

Every technique in the tool came out of the day job at BCN: structured prompting, reliable structured output rather than prose, designing for the awkward cases rather than the easy ones. The client work paid for the education, and the education paid me back with something I use every week — and something I now want to take to market.

What actually transferred from sales?

More than people assume. Sales is discovery — working out what someone genuinely needs when what they’ve asked for is something else. It’s briefing clearly, handling the answer you didn’t want, and going again. That is almost exactly what building with agentic tooling requires.

The syntax was new. The posture wasn’t. The gap was never confidence or curiosity, it was vocabulary — and in this team, vocabulary is the fastest thing to close, because everyone around you is willing to teach.

Where does this go next?

I’m still early, and I know precisely how much I don’t know. The list gets longer rather than shorter, which I’m told is the correct direction.

But the loop is running: client problem in, working thing out, lesson kept. Sometimes the lesson turns into a tool of my own. That loop is the actual asset — and BCN built it with me. I’m heading back into second year knowing the relationship carries on around university, and that when I come out the other side, this is what a career in AI engineering can look like. If you’re a student wondering whether there’s a route into this industry that doesn’t start with a computer science degree: there is. I’m on it.

Thanks Billy – wishing you all the best for your last year at uni – we’ll see you soon!

Billy works with the Client Innovation Team at BCN and is a second-year Business Economics student at Cardiff University. He joined as an undergraduate intern, and BCN has kept the door open for him to continue working and developing with the team around university life.

BCN is an innovative AI solutions company helping organisations put agentic AI to work — and a business committed to developing the next generation of AI engineers.