Writing

Thoughts on AI, engineering and leadership

Essays for the people who build and run software at scale.

A dark control room where a single sealed glowing cube labelled with a unit sits between a stream of raw model tokens on the left and an invoice on the right, with the inside of the cube hidden from view.

AI New 13 October 2026 · 11 min read

Token arbitrage: the AI price gap inside the unit you buy

Token arbitrage is the difference between what you pay for an AI unit and what the outcome costs to deliver. What makes it up, and what to ask.

AI New 8 October 2026 · 7 min read

The AI effect: why AI keeps moving the goalpost

Chess, Go, speech and code stopped counting as AI once they worked. A timeline of the goalpost moving since 1943, and what the pattern says about what is next.

A wide night landscape of a long field with a goalpost standing far in the distance, and a trail of earlier goalposts fading behind the viewer, each marked with a small glowing icon of a past AI milestone such as a chess piece, a Go stone and a speech wave.

Leadership New 25 September 2026 · 16 min read

The missing middle: what AI does to the org chart

The common story is that AI takes entry-level jobs first. My read is that it gets juniors to mid-level in months and gives the middle no shortcut to expert.

A night mountain scene: a bright funicular of blue light, a small glowing orb riding above each glass car, carries people swiftly up the foothills to a wide, mostly empty terrace halfway up. From there a few figures set off along lit paths around the mountain, and climbers with headlamps follow a long switchback trail of amber lanterns to a small group beside a beacon on the summit.

AI17 July 2026 · 11 min read

The great AI lockdown has begun

Strava, Garmin and Reddit are closing the free doors. SAP, Salesforce and ServiceNow are metering theirs. AI broke the open-API bargain. Access is repricing.

A dark wall of glowing API doorways; most are shuttered behind padlocks while a few have converted to toll gates with meters, and a queue of small AI agents waits at the barrier.

AI1 July 2026 · 15 min read

AI is either as good as it gets, or the worst it'll ever be

The AI debate fixates on capability. The variable that decides whether AI earns its place in what you build is cost, and falling cost is the base case.

A dark high-rise office at dusk: a laptop shows an LLM inference-cost chart falling from $60 to $0.06 per million tokens, beside a whiteboard summarising the cost-versus-capability argument.

Leadership18 June 2026 · 16 min read

Enterprise support was built for a clock speed that no longer exists

AI sped up exploits, vendor releases and agent workflows at once. The support model built for annual upgrades cannot keep up. A wake-up call for app leaders.

Three clocks side by side running at different speeds, representing the exploit clock, the vendor-release clock and the AI-agent clock all converging on enterprise application support.

AI5 June 2026 · 9 min read

AI isn't killing content, it's exploding it

Surely AI should mean fewer documents, apps, and sites. Like email and the paperless office before it, it is multiplying them instead.

Leadership18 May 2026 · 17 min read

Vibe Coding in the Enterprise: Why Citizen Developers Are the Real Opportunity

How CIOs should think about vibe coding for citizen developers. Governed prototyping on a sanctioned platform, with a clear graduation path to engineering.

Engineering5 May 2026 · 9 min read

95% of companies get no ROI from AI. Here is what separates the 5%

The MIT and McKinsey data on AI underperformance points to a consistent pattern. The gap is not about tooling.

Engineering9 April 2026 · 14 min read

Building an MCP server for persistent workout memory in Pelaris

How I built a production MCP server to give Claude and ChatGPT persistent workout memory, and what broke along the way.

Engineering19 March 2026 · 9 min read

I built an AI-powered CI/CD pipeline that manages itself

From user feedback to deployed code. Here's the architecture, the tools, and what building AI at enterprise scale taught me about doing it for Pelaris.

AI10 March 2026 · 8 min read

The interface is the product - AI revolution is a UI/UX revolution

AI capability has outpaced human ability to use it. The bottleneck in 2026 is no longer the model, it's the interaction layer.