Writing archive
Older essays on agentic systems, governance, operations, incentives, and the work of making AI useful in real organisations.
Agentic architecture is not one clever model or one workflow. It is a layered environment of servers, access, runtimes, data, protocols, identity, firewalls, applications, work rooms, visualisation, and operational resilience.
A practical reflection on why LLM firewalls, tool gates, and deterministic controls are really an old Stoic idea: not every thought deserves to become an action.
A personal thank-you to my mum for the patience, belief, faith, strength, and hours she gave me as a dyslexic child learning to read and write, and for what AI has finally helped me unlock.
A researched operational resilience piece on the abrupt suspension of access to Anthropic Fable 5 and Mythos 5, and what it means for European organisations relying on critical US software and AI services.
A simple CEO guide to trying AI agents: run one useful demo, keep data bounded, back up the working state, and start with one repeatable workflow.
A reflective thought piece on moving from rigid software wizards and menu trees toward agentic tools that can work from intent, while still respecting the moments where precision matters.
A short observation from a Stockholm restaurant on why small businesses should make menus, specials, local context, and choices easy for personal AI agents to understand.
A reflective essay on what happens when personal agentic systems make public knowledge, local government information, product evidence, and hidden context understandable to ordinary citizens.
A research-backed reflection on how agentic systems should challenge humans when they believe a decision is wrong: respectful escalation, evidence, thresholds, stop-line rules, and human override.
A follow-on to the agentic-worker education piece, arguing that management should become a general literacy: setting outcomes, judging quality, managing attention, using AI well, and taking responsibility for decisions.
A reflective, research-backed article arguing that the next school skill is not only AI literacy, but learning how to manage interrupting agentic workers: context switching, delegation, judgement, and knowing when to pause before accepting a recommended action.
A practical reflection on using Orchistra and agentic workers to cover the skills a small consulting business needs, while treating token spend as operating capital that has to be watched.
A gentle argument for using more precise AI language: chatbot, generative AI, agentic system, local AI, foundation model, AI system, or safety-critical AI, depending on what we actually mean.
AI is changing product-to-engineering ratios, but the deeper shift is capability: product managers now need engineering-grade judgement, and senior engineers may be closer to the new product role than many expect.
A reflection on agentic AI, productivity, and the backlog of work society has avoided because there has never been enough time.
A reflection from teaching the University of Exeter MSc Financial Technology hackathon: clearer outcomes, examples of good work, tasks, roles, mindset, and room to think help humans as much as agents.
A respectful request to EU policymakers: keep the safety and rights goals, but make practical AI adoption easier through sovereign options, regulatory sandboxes, and clear approval routes for ordinary companies.
A thoughtful opinion piece on how higher employment costs can unintentionally speed up AI adoption, and why the progressive answer is not to fight automation but to make work more productive, humane, and worth hiring for.
Governance slop is what happens when boards mistake policies, assumptions, and half-read papers for actual oversight. An opinion piece on the gap between governance theatre and governing.
For companies whose old systems are a real differentiator, the first AI move should not be to touch the core. Start with small, separated, read-only edge POCs that earn trust before expanding.
A public invitation to people working on digital twins, A2A, MCP, and agent-to-agent communication to compare notes and run small public-safe experiments.
AI and agentic proofs of concept fail when they are treated like normal IT projects instead of onboarding a new kind of team member.
AI hallucination often starts with a bad brief: missing context, missing source links, and no safe moment to ask.
A practical note on how marketing changes when AI agents, not only humans, are researching products and making shortlists.