Manchester Tech Festival 2025: Notes from Victoria Baths

Victoria Baths is one of those venues that rewards the decision to hold a tech conference in it rather than somewhere generic. The Grade II*-listed Edwardian swimming baths in Chorlton-on-Medlock (restored, repurposed, the original pool tanks doing various things that are not swimming) has a specific atmosphere that makes you think slightly differently than you would in a glass box conference centre. I don't know exactly why this is. Perhaps it's the combination of Victorian engineering ambition and contemporary tech ambition sitting in the same room. Perhaps it's the acoustics. Probably it's both.

Manchester Tech Festival's 2025 core conference spent two days there on 24-25 September, before the dedicated AI conference moved to DiSH MCR on 1 October. I went to both, or most of both. The scheduling density of MTF always means you're making choices and missing things you'd have wanted to see. I was there with a few colleagues from N Brown, which made for a useful sounding board across the two days: that particular mix of someone nodding along because they've just solved the problem being described on stage, and someone else frowning because they haven't, is a fairly reliable indicator of whether a talk is landing on something real.

The Keynotes

The headliner talks were anchored by two chief data officers, which says a lot about where the technology conversation is in 2025. Ron Thompson, former Chief Data Officer at NASA, flew in from Washington D.C. to talk about data strategy, AI, and innovation at scale. Orlando Machado, former Chief Data Officer at LEGO Group, brought a practitioner perspective on how one of the world's most beloved brands is navigating the data and AI landscape.

The NASA keynote was, as you'd expect, a slightly specific entry point into the AI in business conversation. The data scales involved, the operational constraints, and the failure-is-not-an-option culture are all quite particular to space agencies. What translated well: the argument that good data strategy and good AI strategy are the same strategy, that the quality of your data is the quality of your intelligence, and that organisations which treat data as an operational afterthought get AI outputs that reflect that. Not revolutionary thinking, but well-evidenced and presented with appropriate weight.

The LEGO session was more immediately applicable to the commerce technology audience. Consider the data challenge: how does a 93-year-old toy company with billions of pieces in existence manage the task of knowing its products, its customers, and the relationship between them? Machado's framing of that problem resonated with a room full of people working on similar challenges at different scales. The LEGO customer data problem is actually quite complex: sets that span decades, a collector market as well as a children's market, a global brand where personalisation has to navigate very different cultural contexts. And it's a company that cares deeply about brand integrity in a way that shapes what AI applications it will and won't pursue.

I had the pleasure of a brief conversation with Ron Thompson afterwards, once the formal programme had wound down and the room had thinned out to the kind of number where you can actually hear someone talking. Over a beer, I asked him what surprised him most about how non-NASA organisations were approaching AI governance. His answer was more interesting than it might sound: most organisations were surprised AI had opinions. The point being that when a system surfaces a recommendation that seems to conflict with what the humans in the room believe, the instinct is often to override the model rather than interrogate either the data or the belief. From someone who'd spent years in environments where overriding a system without understanding it properly is genuinely dangerous, that observation landed with a bit more weight than the usual conference wisdom.

Talks Worth Noting

Matt Squire, CTO at Fuzzy Labs (Manchester's own MLOps consultancy), gave a talk that I thought was one of the more honest provocations of the two days. The title was something along the lines of "Are We the Last Programmers? AI and the Future of Code", which could easily have been the setup for thirty minutes of breathless AI maximalism. It wasn't. Squire's framing was more measured: yes, AI is transforming what programming looks like, yes the junior-developer pipeline is already changing, but the question of whether AI replaces programmers or just transforms what programming means is nowhere near resolved. Anyone who tells you with confidence that they know the answer is selling something.

What I found valuable was the grounding in what Fuzzy Labs actually does. Helping organisations put AI into production means they see the gap between "an AI can write this code" and "this code is in production, tested, maintained, and not causing problems at 2am" more clearly than most. That gap is still substantial. It may be shrinking. It is not gone.

The other session that stayed with me was on the rather more serious end of the spectrum. Kirsty Bradley-Law, who works at DISC Manchester and brings a background in education and SEND provision to digital safeguarding questions, was part of a panel on the developer's role in child online safety. The panel explored the UK Online Safety Act and how responsibility for protecting younger users is increasingly embedded in product design decisions rather than policy documents. The argument that developers are, in practice, the last line of defence between harmful experiences and the young people who encounter them online is not comfortable, but it's accurate. The panel didn't pull its punches.

Kirsty was kind enough to talk to me for a while afterwards on exactly this topic. I was keen to pick her brains on the online safety implications for retail platforms specifically, where the distinction between "commerce" and "content" is increasingly blurred. Her perspective, grounded in years of working with young people in educational contexts before moving into the digital space, was clarifying in the way that happens when someone has actually thought carefully about a problem rather than recently discovered it. Some of those conversations from the corridors and post-session networking are honestly where the value of these events lives.

The AI Conference

The dedicated AI day at DiSH had the atmosphere of people who had moved past the "should we take AI seriously?" question and were working on harder ones. The Booking.com session on AI for FinOps and cloud migration was a useful technical deep-dive into AI in infrastructure contexts rather than product contexts: anomaly detection, forecasting, and workload optimisation in a platform running at Booking.com scale. The lesson that keeps emerging from these talks: the value of AI in large-scale operations often comes before the consumer-facing layer, in the infrastructure that makes the consumer experience possible.

The ethics and accessibility session ("AI for All: Ethics, Accessibility and Building an Inclusive Future") was genuinely interesting, and not in a box-ticking way. The Northern tech community has a track record of taking these conversations seriously rather than giving them a polite 40-minute slot and moving on. The discussion of how AI personalisation systems can inadvertently encode and amplify accessibility barriers, by optimising for majority-population engagement patterns that don't serve disabled or neurodiverse users, was specific enough to be actionable rather than theoretical.

The Year-on-Year Change

Comparing to my notes from MTF 2024, the shift is visible. A year ago, the AI sessions were attended by people who were figuring out what to think about AI. This year, the audience had opinions: sometimes strongly held, occasionally conflicting, and always grounded in some kind of real experience rather than theoretical position. The "I tried this and here's what happened" register was much more prevalent.

The split I noticed was not between enthusiasts and sceptics but between people who'd had good early AI deployments and people who'd had difficult ones. These groups have quite different takes on what to prioritise next, and the difference in their experiences is usually traceable to data quality, organisational readiness, or deployment design decisions, rather than to fundamental questions about whether AI works.

That's a healthy and productive kind of disagreement to be having. It means the community is past the "are we in?" question and genuinely wrestling with the "how do we do this well?" question.

One conversation I had in the networking session after the core conference on the 25th stuck with me. Someone who leads a mid-sized Northern digital agency summarised their experience of the past 18 months in a sentence: "The tools got better faster than our clients got ready for them." That feels like an accurate characterisation of where the market is in early October 2025.