Ancient Rooms, Modern Questions: A Day at the Cambridge Summit 2026
Sitting in a room that has been part of Cambridge University since the 14th century, eating lunch between sessions on AI, headless CMS and behavioural science, I found myself wondering what conversations had taken place in that same room across the centuries. When the railways arrived. During the industrial revolution. When computers changed everything, and then the internet after that. Each era must have had its own version of this question: what does this mean for how we work, and what do we stand to lose or gain?
That contrast between ancient setting and very modern questions felt fitting for a day that asked us to think carefully rather than react quickly.
![]()
The Cambridge Summit is an annual event organised by Boye & Co, bringing together around 30 people from both higher education and the private sector for a day of honest conversation, learning and networking. This year’s event was held at The Old Schools, and I was glad to be there.
Here is what stood out for me.
What I learned last week: PHP, security and more
Presented by Simon Jones, Studio 24
This was the second session of the day, and one I found particularly valuable. It centred on a pair of critical PHP vulnerabilities known as PHP2Shell (CVE-2026-109618 and CVE-2026-109619), each rated 9.2 out of 10 in severity. PHP powers roughly 70% of websites worldwide, including WordPress and Drupal, which between them account for a significant proportion of the web, so the exposure was enormous.
What made the story particularly interesting was the history behind it. Signs of one of the flaws had been identified as far back as 2024, but at the time no concrete attack path could be demonstrated and the fix was postponed due to compatibility concerns and limited team resources. It was not until August 2026 that a researcher used an AI model to confirm that unauthenticated remote code execution was possible. OpenAI submitted a separate report on the same issue the same day. Within weeks, exploit attempts were being recorded in the millions.
Patches were released on 21 October 2026. The FBI and the UK’s NCSC both advised that if an immediate update was not possible, organisations should take affected systems offline. Several countries, including the Netherlands, Finland, France, Germany and Belgium, did exactly that, shutting down web services until it was safe to bring them back up.
What struck me about this session was not just the technical detail but the candid account of how it was handled, and in some cases not handled well. It was a reminder that security is not just a technical issue; it is a communication and decision-making one too, and that known risks left unaddressed have a way of returning under worse circumstances.
Making decisions like a leader: a few words on behavioural science
Presented by Nina Dyrberg, BeWaviour
![]()
This was the session that stuck with me most. I went in expecting a familiar leadership talk and came out with a genuinely different way of thinking about how I make decisions.
The session was built around two decision traps that leaders fall into repeatedly. The first is moving too fast: making decisions on the wrong assumptions, falling into confirmation bias where we test an idea with a lens already angled to confirm what we think, or falling into the binary trap where a decision feels like it can only go one of two ways when in reality there are more options available. The second is moving too slow: the status quo bias that keeps us doing what we have always done, choice overload that causes us to stall when there are too many options, and the sunk cost fallacy, the tendency to keep investing in something simply because we have already invested so much.
One practical tool introduced was a matrix for working out what we actually know before making a decision: known knowns (facts we can use), unknown knowns (hidden knowledge we should trust), known unknowns (assumptions we need to investigate), and unknown unknowns (blind spots where we need to get inputs from others). Simple, but genuinely useful for working out why a particular decision feels harder than it should.
The session also addressed procrastination directly. The model was stark: harsh self-talk leads to guilt, which leads to more procrastination. The alternative, self-forgiveness, leads to fewer negative emotions and greater motivation to start again. As a leader, that is easy to overlook both in yourself and in the people you work with.
The slide that has stayed with me most was this:
“If you believe there is one right decision, you become afraid of choosing the wrong one. But if you see decisions as experiments, there is no wrong choice, only something to learn from.”
The practical takeaways were: ask what you actually know, forgive yourself when procrastinating, identify the smallest next step, ask what your future self would do, and plan the decision rather than just making it.
Open health informatics for the digital transformation of healthcare in Ukraine
Presented by John Chelsom, Seven Informatics
This was one of the most striking sessions of the day. It covered how open-source software is being used to support remote healthcare delivery in Ukraine, in extraordinarily difficult circumstances.
It is a powerful example of what open source can do when it is applied with purpose. EdWeb 2 and MyEd are both built on open-source platforms, and hearing how the same principles are being used to deliver critical healthcare services put our own work in a different perspective. It was humbling and motivating in equal measure.
Using AI to understand our work better
Presented by Ben Lee, Instruction
![]()
This session opened with a brief history lesson: the 2017 Google paper “Attention Is All You Need”, which introduced the Transformer architecture behind ChatGPT. The key insight from that paper is elegant: every word looks at every other word to understand context. The example used was “bank by the river”, where the word “bank” looks to “river” to know which kind of bank it is. A useful reminder that what feels like intelligence is fundamentally about context.
From there, the session moved to a quote from Greg Isenberg (October 2026) worth sitting with: “We’re maybe 12 months from an agent that watches you work for a week and then does your job.” The slide alongside it made the crucial distinction: AI can see what you do, the steps, the tools, the order, the clicks. It cannot see why. It misses your reasons, what you checked, the times you said “no, that’s wrong.”
That distinction is at the heart of what the session argued: capturing expert judgement becomes the valuable bit. The practical example was building AI agents for pitch review, where the expert’s judgment was turned into a rubric, used to train an assessor agent, and pitch win rate improved from 40% to 60% in a year. The process was clear: understand the work first (interview the expert), write the playbook, validate it, build, test, check, then use it for real. Four of those seven stages require the expert.
The message I took away was that the right question is not “what can AI do?” but “what does the expert know that AI cannot see, and how do we capture it?” Before any agent is built, the session argued for a 45-minute interview covering how the work really runs: the rhythms, the decisions, the dependencies, the knowledge nobody ever wrote down, and where the work hurts. As the slide put it: “You can’t describe work you’ve never been asked to explain.”
This connects to something I keep returning to. If organisations use AI to remove the junior roles that build expertise over time, they also remove the pipeline for creating the people who can answer those questions in the future. The expert judgement that makes AI useful must come from somewhere.
Super, intelligent layouts: headless the way it was always intended
Presented by Markus Schork
![]()
The session on headless CMS and intelligent layouts was directly relevant to EdWeb 2 and the future of MyEd portal. The core idea was that layouts should adapt dynamically based on the content that is added, rather than being fixed templates that content is forced to fit.
EdWeb 2 is already designed with this principle in mind to a degree, but this session made me think about how much further we could take it. Making layouts more responsive to what users actually input is something I want to explore further with the team.
Beyond the sessions
![]()
After the formal sessions, we had lunch in a room that has been part of Cambridge University since the 14th century. It is worth taking a moment to appreciate where you are on days like this.
After the event, I joined colleagues from the Cambridge web team at The Eagle, the pub where Watson and Crick famously announced the discovery of DNA. Good conversations continued with the Cambridge University web teams.
The following day I had productive meetings with the web teams at Oxford, which alongside the conversations at the summit gave me a useful picture of how peer institutions are thinking about the development of their web services.
What I am taking away
Three things from this year:
- Decision-making as a practice. The behavioural science session has given me a more conscious framework for handling difficult decisions, and I want to bring some of that thinking into how we work as a section.
- Open source has reach beyond our sector. Seeing it applied in Ukrainian healthcare is a reminder of why the open-source model matters and why we should continue to champion it.
- Ask harder questions about AI. Not whether to use it, but how, and at what cost, including the human cost of removing the roles and pipelines that build the expertise AI depends on.
Back in that room that has looked over Cambridge since the 14th century, I doubt the questions were so different. How do we use what is new without losing what matters? Worth thinking about.
If you are considering attending an event like this, my view is that the value is not just in the sessions but in the conversations between them, and the perspective you get from stepping outside the day-to-day.