I go to tech conferences for one reason: to understand how the world of work is changing around us, before it changes on us.
This year I was at both Dublin Tech Summit and London Tech Week. The themes overlapped more than I expected. Both events had a heavy focus on AI, that is not a surprise at this stage, but what was notable was where the focus sat. Not on the technology itself, but on the human and organisational problems that come with it.
These are my notes from two panels at London Tech Week that I thought were worth writing up.
Security risk and the governance gap
The panel "Ensuring Growing AI Use Isn't Increasing Security Risk" brought together security leads from GSK, the UK Department for Education, the Responsible AI Institute, and Faculty AI. The conversation covered ground I think is underappreciated outside specialist security circles.
The first thing worth noting is that AI risk is not one thing. The panel drew a clear line between safety risk (bias, inaccuracy, traceability), operational risk, and cybersecurity risk (can the system be compromised or turned against you). These need to be treated differently, even if they need to be managed in coordination. You can have a model that is unsafe but well-secured, or a secure model that gives you dangerously wrong outputs. Conflating the two leads to poor decisions about where to invest.
The more interesting part of the conversation was about governance. The argument made, and I think it is correct, is that the real risk is not AI adoption itself, but adoption outpacing governance. The car brakes analogy landed well: brakes do not slow a car down in some absolute sense, they enable the driver to go faster because they can manage the risk. Governance done well works the same way. It enables velocity rather than killing it.
Where governance tends to fail is when it is added after the fact, or when it is designed to look like governance rather than to work like it. With agentic AI in particular, systems that make decisions and take actions autonomously, you cannot retrospectively audit what an agent did if you did not build the traceability in from the start.
The shadow AI point is one that people in organisations should take seriously. Martin Sivorn, CISO at the UK Department for Education, put it plainly: they never solved shadow IT and they are not going to start by solving shadow AI. Shadow AI is a demand signal. It means the officially sanctioned tools are not doing the job. If they were, people would use them.
The implication for organisations is straightforward: if you want people to use your approved tools, the approved tools need to be good enough. A sandbox environment with corporate data properly contained is more effective than restrictions that push people toward uncontrolled alternatives. Lack of consequences for unsanctioned AI use also drives the problem. People need to understand what is and is not appropriate, and there have to be repercussions when people who know better do it anyway.
On data: the point that AI exposes your data is not something you can engineer around entirely. The question is whether you have the right controls calibrated to the actual risk level. An internal translation tool is not the same risk profile as a customer-facing AI system. Applying the same controls to both is not careful governance; it is a drag that pushes people toward workarounds.
The skills question
The second panel, "The Upskilling Imperative: Preparing for the Jobs of Tomorrow", took a different angle. The conversation was less about security and more about what this period actually requires of workers and organisations.
The framing that stuck with me was from Professor Geoff Rodgers of Brunel University: AI is going to eradicate some jobs, create some new ones, and transform a large number of jobs. That third category is the one most organisations are not preparing for adequately.
The panel made the case that technical AI literacy is necessary but not sufficient. What organisations actually need, and what is genuinely difficult to hire for, is people who combine a working knowledge of AI with strong human judgement. Curiosity, critical thinking, empathy, human-centred leadership. The problem, as Gori Yahaya of UpSkill Universe put it, is that nobody puts curiosity on their CV. The conventional hiring process is not designed to surface these qualities, and training programmes are not consistently designed to develop them.
The broader point from the panel was that the question is not "AI or humans". It is about understanding what AI does well and what humans do well, and building systems and teams that use both appropriately. Human in the loop is not enough; human in the lead is the aim.
For organisations thinking about talent strategy, the German post-war model came up as a reference point: invest in skills at a local and sectoral level rather than relying on the labour market to provide. The UK has Brexit as a structural complication here that Ireland does not, but the underlying point applies: waiting for the right person to appear is not a strategy when the skills you need are still being defined.
There are training opportunities available to Irish professionals worth knowing about if upskilling is something you are thinking about for yourself or your team.
What both events had in common
Dublin Tech Summit covered similar ground on the security side. The emphasis there was also on how to secure AI systems, a theme that is clearly not going away.
What struck me across both events is that the most useful conversations were not about AI capabilities. They were about the gap between where AI development is and where AI governance, insurance, and skills development are. That gap is real and it is growing. The organisations and individuals that close it faster will be better placed than those that treat it as someone else's problem.
If you want to talk through how AI is affecting your sector or your work, email sarah@dedico.ie.
Frequently asked questions
What is shadow AI?
Shadow AI refers to the unsanctioned use of AI tools by employees outside officially approved platforms. It typically signals that the tools an organisation provides are not meeting staff needs.
What is the difference between AI safety and AI cybersecurity?
AI safety covers issues like bias, inaccuracy, and traceability, things that make a model unreliable or harmful in operation. AI cybersecurity is about whether a model can be compromised, hacked, or turned against its users. They are distinct but need to be managed in close coordination.
What skills will workers need in an AI-driven economy?
Technical AI literacy matters, but panellists at London Tech Week 2026 argued equally for judgement, curiosity, critical thinking, and human-centred leadership, the skills that AI cannot replicate.