Blog
Thoughts on AI, technology, IT service management, and digital transformation.
I Run Three Sprints a Week. I'm Not Sure They're Sprints.
AI dropped a two-to-three-week bug fix to two-to-three hours, and the word 'sprint' stopped fitting the work. Why the real job of the ceremony has shifted from pacing delivery to protecting the decision.
The Moat Was Never the Code
For decades, the ability to write code was the moat. AI just drained it. Here's what's actually hard now, and which engineers get left behind.
The Gap I Didn't See Coming
A week in Copenhagen, four sessions, an audience just shy of a thousand people, and the question I came home with: the real one for vertical software businesses right now isn't whether AI matters. It's how soon we let our people build.
Beyond Basic RAG: How HyDE and Cohere Reranking Changed Everything About My Pipeline's Output Quality
The right chunk was surfacing in the top results roughly half the time on real-world, conversational queries. After adding HyDE and Cohere reranking, I'm consistently above 90%. In short: HyDE makes your search query look more like the answer you're hunting for, and reranking reorders what comes back by how well it actually answers the question. Here's why that gap exists, and how these two techniques close it.
The Founder in the Corp: How Entrepreneurial Thinking Survives Inside a Software Company
I co-founded a business from £5k to £15M, exited, and stepped into a corporate tech leadership role. Here's what I've learned about keeping the founder instinct alive inside a structure that wasn't built for it, and why stopping pretending it can be switched off was the best decision I made.
Why Your Ego is Breaking Your AI Strategy
The loudest voices on AI are either predicting doom or dismissing it as "spicy autocomplete", and neither is building anything. The people who ship share one trait: low ego. But ego isn't stupidity, it's protecting something real, and that's why low ego is so rare and so valuable for AI strategy.
The Day I Realised AI Has a "Try Hard" Mode (And How to Trigger It)
Context isn't there to make output sound nicer. It tells the model how much scrutiny the task deserves. Add identity, audience and stakes, and the same model reasons at a different level entirely.
Success This Year Had Very Little To Do With Technical Skills
What stands out from a strong year in tech isn't what was achieved, but the behavioural engine that drove it. Seven core traits shaped sustained, high-value success in 2025: curiosity, courage, persistence, integrity and the habits behind them.
Set Free: How AI Unlocked a Lifetime of Ideas
2025 has been transformative. For the first time in my career, I feel like I've caught up with my own ambition. This is the story of how AI removed the barrier between vision and execution, and why I'm genuinely thriving.
Escaping the Desert of Innovation: How Vibe Coding Liberated this Technical Founder
For years, I was trapped in the "desert of innovation", full of ideas but dependent on others to build them. Vibe Coding changed everything: natural language interaction with LLMs turned me from a frustrated product owner into a fully fledged builder who can execute on vision.
Why Your MCP Servers Are Overcharging Your LLM (and How to Fix It)
MCP servers can burn 15,000+ tokens just loading their definitions, before doing any actual work. Here are the two sources of Context Bloat, static definitions and dynamic results, and the design principles that keep a server lean.
The Customer-First AI Mandate: Solving ITSM's Toughest Pain Points
True AI value in ITSM isn't found by asking "how can we use AI?" but by asking "What is the worst part of our service experience?" Intelligent Triage Engines remove admin time and turn analysts from data-entry specialists into high-value problem solvers.
The Great Pivot: Why ChatGPT Is Moving Away from Reddit's Data Goldmine — and What It Means for AI's Future
ChatGPT is dramatically reducing its use of Reddit content, dropping from 14% to 2% of citations. This signals a shift from volume to quality, with verifiable, licensed data taking priority over unstructured user-generated content in AI and enterprise applications.
Why Your LLM Outputs Are Inconsistent (And How to Fix It)
With GPT-5 now available, the temperature parameter is disappearing from newer models. LLM outputs are inherently inconsistent, so reliable applications need hyper-explicit prompts, robust validation and testing built around unpredictable foundations.
The Quest for Full Context: From Paper Stacks to Intelligent AI Retrieval
From scanning mountains of paper with Kofax Ascent Capture to building AI-powered document intelligence systems, the mission remains the same: make information accessible. Multi-modal document intelligence turns RAG systems from keyword retrievers into genuine problem-solvers.
The Burger King Lesson: The Unquenchable Thirst for Better
A simple observation at Burger King, a dropped drink kicked under the counter, revealed a universal truth: there's always an opportunity to leave things better than you found them. That principle has shaped my career from kitchens to code, and AI tools are helping accelerate everyone's journey.
The Entrepreneurial Spark Never Fades
From refurbishing Mini engines in my parents' garage to leading AI initiatives as a Technical Operations Manager, the entrepreneurial mindset remains constant. That fundamental curiosity, seeing how something works and making it better, drives innovation across every role.
Packing for Performance: How to Tune Your Vector Index This Summer
While you're recharging your batteries this summer, your vector index could use a tune-up too. Move beyond default similarity search and learn how to craft custom scoring profiles that combine semantic search, keyword matching, and freshness factors for truly relevant results.
From a 1541 Floppy Drive to a Core Value: A Lifelong Journey of Curiosity
Awarded "Be Curious" at our Town Hall, I traced my curiosity back to a Commodore 1541 floppy drive and a simple question: "Why not?" That early spark has shaped my career and still drives innovation in modern AI and technical leadership.
My Brain Has Too Many Tabs Open: Squaring My Role as a Tech Leader with My Introverted Reality
I can architect complex cloud infrastructure, but forced small talk sends me into panic. As a Technical Operations Manager, I've learned that the question isn't "How do I fix myself?" but "How do I work with myself?" I'm building a career that honours my introversion rather than fighting it.
The Goldilocks Zone of NLP Search: Why Chunk Size Matters in Your AI Vector
In vector search, chunk size quietly decides whether retrieval returns useful context or beautifully embedded noise. Here's how to size chunks for RAG by content type, how much overlap to use, and when to stop sizing by length and split by meaning instead.
Turning Up or Down the Heat: How Temperature Settings Impact Azure OpenAI Responses
As the mercury soared above 30 degrees, I found myself reflecting on the "temperature" setting in my AI models. This is the parameter that controls how random or creative Azure OpenAI responses are, with practical strategies for tuning models through experimentation and A/B testing.
How a Title Shift Sparked a Paradigm Change
I spent years as a founder thinking job titles were meaningless. Then changing mine from Supervisor to Technical Operations Manager changed how suppliers and vendors dealt with me, and I had to admit I'd been wrong. Why titles are signals that do real work in any room where people don't already know you.
Beyond the Phone Call: How Vector Databases Can Transform Your Business
As a busy contact centre manager, you've all been there: the phone rings off the hook, a steady stream of potential customers calling for service. Vector databases combined with AI can turn customer conversations into intelligence and transform inconsistent sales outcomes into predictable success.