The independent voice for when
AI and security collide.

Every conversation you have with an LLM quietly becomes more expensive, and not in the way most people think. It is not just that you are sending more messages. Each new message replays the entire chat so far. Output tokens cost several times more than input tokens. They compound on every round. By turn ten, you are paying for the same fluff dozens of times over. Enter Caveman: a 70-line Claude Code skill from Julius Brussee that makes Claude talk like a smart caveman. Articles, hedging and pleasantries gone, technical accuracy fully intact. Real-world result: roughly 75% fewer output tokens for the same answer. In 15 minutes I will show how token growth actually works (worse than linear), demo the skill live and unpack why a joke prompt is one of the most effective cost levers I have found this year. Bring your wallet. Leave your articles, adverbs and "I'd be happy to help" at the door.

In 30 years, Niklas Luhmann produced over 70 books and hundreds of academic papers using a system of index cards and boxes he called his "second mind", or Zettelkasten (slip boxes). However, most people who tried to replicate his method gave up within weeks. Recently, his method has had a resurgence in modern AI discussions, because the way we store and access knowledge may be the biggest unfixed bottleneck in our AI stack. This talk is on how the modern version of the Zettelkasten method addresses that: what it is, how to implement it and why you should try it. Spoiler alert: when your AI tools underperform, it might not be because the models are weak. It could be because of how you've stored what you're asking them to think about.

Every UK Lotto combination has the same odds of winning. Or do they? People pick their numbers in predictably biased ways, which changes not your chance of winning but how much you would take home if you did. I used AI as an end-to-end analytical partner to frame the question, build the strategies and test them against more than 1,000 real UK draws from 2016 to 2025. I'll show you what held up, what fell apart and how the same approach applies to any problem where the obvious answer turns out to be only half of it.

Panellist on the "Cybersecurity, Sovereignty and Trust" panel, alongside Dr Viktoria Schmittmann and Dr Jaroslaw Wilk, moderated by Stuart Sumner and covered by CRN and ChannelWeb.

One LinkedIn post hardly gathers traction whilst another goes viral. I wanted to know why, so I used AI to find out. What I found surprised me, and led me to build a system for predicting the likely engagement and reach of any LinkedIn post before it goes live. In this session I'll share how I used AI at every step: to conduct the research, spot unexpected patterns and build the scoring system itself. I'll also give you access to the tool, so you can score and optimise your own posts on the day. Whether you're using LinkedIn to build a personal brand, grow your business profile or drive sales, you'll leave with a new mental model for what makes content perform, a practical tool you can use straight away and a clearer sense of how to approach your own problems with AI. There's a live demo at the end. Bring a draft LinkedIn post.

We've reached an inflection point: stakeholders are scrutinising AI Security posture. Boards are seeking reassurance, procurement teams are probing and regulators are circling, and all want evidence that your AI systems and the data they process are properly governed. This talk covers Shadow AI risks, the governance gaps most organisations miss and practical principles your team can apply to prepare for and evidence AI Security practices.