Key Takeaways from Scottish Web Folk 2026
It was a busy day at the Scottish Web Folk conference, packed with ‘quick-fire’ sessions that lasted just 30 minutes each. The theme was Human vs Machine, however I took away no negative vibes about AI. The general feeling was positivity, and I enjoyed hearing how AI is altering things – how it can enhance our workflows and what it can add to a process with the correct guardrails in place.
Andrew Millar, Head of Digital Engagement & Marketing at the University of Dundee, kicked off the day. His session was thought provoking, and I jotted down several key takeaways that set the tone for the day ahead:
- AI is a gateway to something both amazing and terrifying.
- Just because things have worked out previously regarding innovation, we shouldn’t view that as comfort; we must accept the responsibility that comes with it.
- Machines are now using our content and information to speak on our behalf.
- How has this shift broken our existing metrics?
- The machine has inserted itself into the conversation and is now providing the answers.
- How do we determine if it is actually using our content?
- We are using AI more and more in our daily lives, yet people are trusting it less.
- A new acronym for me: GIGO – Garbage In, Garbage Out.
These points framed the day perfectly, as the subsequent presentations addressed these challenges and more.
![]()
The human side of digital transformation
It was interesting to hear from Mark Kennedy, Web Development Manager at Ulster University, about how he leads his web team of four, who support over 400 publishers. Mark and his team do not sit within Information Services, but instead they are integrated into Marketing and Communications. He mentioned that this can come with some challenges but also fosters more involved collaborations around content. For us in the Web Development Team at the University of Edinburgh, we are somewhat removed from the quality and accuracy of the content our publishers upload to their sites. As a wider team, we focus on providing the tools, training, and advice required to produce good-quality, accessible content. In contrast, at Ulster, the web team works closely with their MarComms colleagues, acting as gatekeepers for good content.
This close working relationship benefitted the team when Mark’s MarComms colleagues requested a sandbox to experiment with the new component-based platform, they built a publish-ready site within just one week. Although intended only as a playground, they fully embraced the reusable components – a surprising yet fantastic outcome that truly illustrated the human side of digital transformation. It struck me how accustomed we become to our own ways of working and the specific colleagues and disciplines we interact with daily. Mark’s talk sparked a thought: what would it be like if we were structured differently?
Post deployment testing
I also attended Rob McCreary’s talk (University of Dundee) on testing. He emphasised the importance of introducing automated testing techniques to enable quick fixes that can be deployed with confidence, specifically detailing how his team uses Behat feature files to confirm the live environment is healthy immediately after deployment. This was quite a complex topic, and while our team is just at the start of introducing automated testing, here are my main takeaways:
- It’s not only developers who ship bugs – integrated content can too.
- Import and integration failures can reach the live page before anyone notices.
- A GitLab pipeline can run a test report and display the results clearly.
- Build in a clean-up process: if entities (pages or users for example) are created as part of a test, they must be cleaned up afterwards.
- Fast smoke tests can provide an early warning system.
- Scheduled regression tests cover the wider system without holding up releases.
Side note: As I write this post, I find myself cross-referencing my notes, reviewing photos of the slides, and looking up terms I didn’t fully grasp during the sessions. This process itself feels like part of the lesson to me.
Thorough scoping with the help of AI
Matthew Jones and Robin Butler from Vepple, a virtual experience platform for universities, presented on using AI to draft development specifications. Their workflow begins with an MRD (Market Requirements Document), which outlines the problem statement and testable acceptance criteria. The most intriguing part of their talk was the concept of a ‘Grilling Session.’
At the time, I jotted down that this should be ‘thorough in a positive way,’ but I wasn’t entirely sure what that looked like. Researching it later revealed that this practice flips the conventional use of AI on its head. Instead of the user simply prompting the AI to generate content, the AI agent takes the lead. It prompts the developers with tailored, iterative questions – effectively ‘grilling’ the user to clarify scope and refine requirements. This rigorous interrogation ensures nothing is missed before the AI outputs a PRD (Product Requirements Document).
As they put it: “The agent writes with us and not for us.”
Once the specification is solid, the agent can slice it into actionable tickets for humans to follow, organising them into epics (projects). They also mentioned that automated tests are used to ‘pin’ existing functionality, ensuring that new changes remain small and manageable.
Takeaway: We don’t keep a platform in good shape by going faster – we do it by being thorough.
![]()
Findability and AI-driven search
Paul Costello and Rob Allen from Numiko talked about their work on findability with the University of Oxford. I know about SEO – it’s been around for some time! – but GEO and AEO were new to me. These stand for Generative Engine Optimisation and Answer Engine Optimisation, which are basically the same thing: looking at how AI-driven search works. This ties neatly back to the points from the introduction.
Honestly, this session was very content-focussed, so it’s not entirely relevant to my daily work as a developer. But I think it’s still really interesting and valuable for us to understand. And actually, it makes a lot of sense:
- Think of a list of non-branded prompts that a user might type into an AI.
- Structure your content around the questions users will actually ask.
- Clear out old, out-of-date content.
- Make sure how you talk about your organisation matches how the audience would phrase it or ask it.
It’s a simple idea – if people are asking AI questions in their own words, our content needs to speak the same language.
Extensive web estates and the importance of governance
To finish the day, I sat in on an unscripted presentation by Stratos Filalithis (University of Edinburgh) and Kris Purdy (University of Glasgow). There were no slides, just an open conversation between them and the audience about governing their massive web estates. It’s incredible just how large these estates are and I was genuinely surprised when Stratos quoted around 2,000 sites.
It’s a huge task – finding these sites, documenting them, establishing if they’re still required, tracking down owners, checking where they’re hosted, and essentially (fitting perfectly with the theme of the day) auditing the content.
The scary part? Sites from the late nineties could still be out there, serving AI agents with outdated answers! Eek!!!
Conclusion
Looking back at the day, the common thread wasn’t about fearing AI but was much more positive – it was about how we prepare our content and how we can employ agents to help with workflows, development, scoping to name just a few use cases.
Quality matters more than speed.
If we don’t audit our estates, we risk those forgotten sites from the late nineties feeding bad information to AI agents. And if we don’t speak the same language as our users, AI won’t find our answers anyway.
The landscape will keep changing, but if we focus on getting the basics right, we’ll be ready for whatever comes next.