AI is changing how quickly software can be built. But faster doesn’t automatically mean better.
At Vepple, we’re a dedicated team building and maintaining an immersive virtual experience platform trusted by universities across the UK. As we work on the platform, we’ve been looking closely at how we build software, where AI fits into our process, and, just as importantly, where it doesn’t.
Because the real measure of good technology isn’t how quickly it’s built. It’s whether it makes someone’s experience better.
For us, that means helping prospective students build the bigger picture, create meaningful moments and connect emotionally with their next steps. Our platform can’t just work in theory. It has to work for real people making life-defining decisions.
We’ve tested several ways of working with AI over the past year. Some tests worked brilliantly, whilst others… not so much! But these were still useful as they showed us exactly what not to do.
Our experiments are primarily focused on software development, but the lessons can be relevant to anyone using AI in their work, including university marketers. How do we make the most of what AI can do without losing sight of the people we’re doing it for?
We spoke to Head of Product, Matt Jones, to get some insights and tips for university marketers.
AI can build quickly. But does it understand what you’re building?
“Our first experiments with AI were fairly straightforward,” explains Matt Jones, Head of Product. “We used it to write code snippets, answer technical questions, and speed up product documentation. On the whole, these initial tests worked well and helped our team automate repetitive, time-consuming tasks.”
“So we pushed further,” Matt continues. “We asked ourselves: what if we handed an AI agent a feature request and let it build the entire solution independently? Could it handle that, and would it deliver high-quality work?”
“We gave an AI agent a ticket to tidy up the tour footer in our Guided Tours interface,” says Matt. “The agent adjusted the footer as requested, and all automated tests passed without error. However, there was a noticeable flaw: the Ambassador Bubble expanded slightly, breaking the intended halo effect and creating two overlapping circles instead.”


Obviously, one of our team caught the visual bug during manual testing, so it never reached production. But the risk was clear.
The AI agent had followed the instruction, but it hadn’t understood the wider experience it was contributing to. It could check whether the code worked, but it couldn’t reliably judge whether the result made sense for the student using it.
Marketers face the exact same risk when using GenAI to write course pages, email nurturing sequences, prospectus blurbs, or even artwork. An AI tool can draft a course overview in seconds. The automated checks pass: the facts are accurate, the grammar is sound, and every target keyword is present. But completing the task isn’t the same as creating an emotional connection. If prospective students read a profile that feels sterile, they simply can’t picture themselves living and studying on your campus. AI can build the structure, but only people understand what makes an applicant feel at home.
Too much process isn’t the answer either
“Our next experiment went to the opposite extreme,” explains Matt. “Instead of letting AI build features directly, we invited AI to manage and support our core processes. We trialled an AI development framework that deployed multiple specialist agents to manage every project phase, from product management and architecture to code testing,” Matt recalls.
The issue was, for a single internal feature, the system generated 55 separate documents containing 113,000 words. That’s longer than an average novel!
We’d created a process that was supposed to make our team’s lives easier, but instead it was getting in the way of the people doing the work. More documentation didn’t mean better decisions, and more AI activity didn’t mean better outcomes.
University marketing teams know this trap all too well. When teams are stretched and deadlines loom, AI shouldn’t be used to churn out endless strategies or audits just because it’s quick to produce. Generating more documents doesn’t lead to better decisions, and if nobody has time to read them, they achieve nothing.
Instead, use AI to do the opposite: let it summarise dense feedback, help pull key insights from messy data, or condense long reports into clear next steps. By using AI to strip back the bloat rather than add to it, your team gets time back to focus on what actually moves the dial for applicants.
Finding the balance
Neither extreme was working for us. We didn’t want AI making decisions without understanding the people affected by them, but we didn’t want to create so much process that our own team couldn’t get anything done.
“So we redesigned our workflow around collaboration,” explains Matt. “We now use AI to help us think, challenge assumptions and plan, while keeping our engineers and product specialists firmly in the driving seat.”
The aim is simple: let AI take on the work it’s good at, while keeping human judgement where it matters most. Here is how that works in practice:
1. Challenge the idea before building it
“We begin by using AI to interrogate our initial feature ideas,” says Matt. “We ask what edge cases we haven’t considered, where user friction could occur, and what happens in non-standard scenarios.”
But AI needs context to ask the right questions. So, we also use it to record our product discussions and feed those transcripts back into the process. That way, the agent isn’t just working from a list of requirements. It has the wider context about the challenges we’re trying to solve, the decisions we’ve made and the people we’re solving them for, as well as all the back and forth conversation between our teams in between that led to the decision being made.
For university marketers, interrogating your work isn’t simply saying “proof this copy”. It’s about asking tough, empathetic questions: What does this open day page look like to a first-generation applicant? Where does an international student feel friction when checking accommodation costs? Using AI to challenge your own assumptions helps you design for empathy before a campaign ever goes live.
2. Agree what “good” (and what “done”) looks like
“We use open team discussions to establish a single, clear product specification, supported by an AI tool that understands our current codebase,” Matt notes. “Our team reviews the specification, questions assumptions and sets explicit technical boundaries.”
Crucially, we define what “done” actually looks like, including how the feature should work for the person using it. AI can execute choices rapidly, but deciding which choices are right requires an understanding of the product, its users and the experience we’re trying to create.
For recruitment teams, the trap is identical. If you ask an AI tool to generate an applicant nurturing sequence without defining clear boundaries (what tone represents your university, what specific anxiety you’re addressing, and what action you want the student to take) you end up with polite, forgettable noise. Before writing any prompt, agree with your team on what “done” actually looks like for the student on the other side of the screen.
3. Make it visual before it becomes code
“Before we build any customer-facing changes, we produce an interactive prototype for internal review,” explains Matt. “This allows colleagues across Product, Engineering, Marketing, Sales and Customer Success to interact with the proposed feature directly.”
You don’t need to decipher a complex technical brief to spot when a user journey feels counter-intuitive or when a visual component fails to reflect our standards. More importantly, it means we’re getting different perspectives involved before a decision becomes harder to change. Everyone has an opportunity to consider how the experience will work for the people who ultimately use it.
Let’s take website copy as an example. Too often, teams spend weeks debating copy in spreadsheets before anyone actually sees what the student will experience. You can use AI to break that cycle. Instead of passing flat text briefs back and forth across faculties, use generative design tools to spin up a visual, working mockup of your landing page or email sequence in minutes. When faculty heads can interact with a tangible draft on day one, feedback changes instantly. You stop arguing about individual adjectives in isolation and start spotting where the visual journey actually falls down for an applicant. The visual flow is just as important as the words on the page.
4. Keep progress visible with rigorous quality checks
“Once the team approves the prototype, we break the build into focused, bite-sized tasks with checkpoints at three milestones,” says Matt.
- Before building: We explicitly define the recruitment challenge we’re solving, the functional acceptance criteria and what ‘done’ looks like for the end user
- During building: We protect the rest of the platform, maintain security and have a human checking progress throughout
- Before release: Someone other than the original builder reviews the code. Product then signs off the user experience, and customer-facing documentation is finalised.
These checks uphold the standards we’ve always worked to, making sure that introducing AI into our development process doesn’t mean compromising on the quality, reliability or experience our university customers expect.
The same discipline protects recruitment campaigns. When speed increases, governance often breaks. Having structured milestones, where someone other than the prompt author checks the emotional resonance, accessibility, and brand tone, ensures that rapid AI output never dilutes your institutional reputation.
Why we’re doing things differently
“We’re continually refining our approach as new tools emerge. But our fundamental principle remains the same: technology only works when guided by the people who understand the people using it,” Matt concludes, “For us, that’s prospective students making important decisions about their future.”
The message Matt wants teams to take away is this: It’s easy to get caught up in what AI makes possible. The new tools, the efficiencies, the things we can now build that would have taken weeks before. But possibility isn’t the same as purpose.
The most important question isn’t how much more we can do with AI. It’s whether what we’re doing is making a meaningful difference to the people on the other side of it.
- Looking to build deeper digital connections with your prospective students? Explore our interactive demo to see how human-led, immersive experiences bring campus life to students worldwide
- Missed our session at Scottish Web Folk? Take a look at our curated list of AI resources for higher education teams


