Job card № 001 · opened Jun 2026
Practical AI from the conference stage
workshop floor.
I repair vehicles, ECUs, diagnostic workflows — and AI systems. These are field notes from building tools that have to work with real customers, real hardware, real servers and real constraints. Not theory. Not hype. Things that either worked or didn't, and what I learned either way.
Latest from the bench
All notes →The Human Brain Was Not Built for the Infinite Feed
John's field note on information overload, family, trust, local AI, and why private AI systems are not about hiding — they are about keeping human agency alive.
12 June 2026 · 12 min readBody-Adjacent AI Needs a Kill Switch on Our Side
If AI is moving into glasses, cars, tools, workshops and eventually body-adjacent interfaces, the off switch cannot live in the cloud.
Field-testedFable 5 Feels Like Two Models in One
An operator's observation from tool-heavy sessions with Claude Fable 5: it behaves less like a single plain model and more like a front controller in front of a reasoning engine — and that pattern is exactly how we should be building our own agent systems.
Field-testedWhy I Rebuilt My AI Agent’s Memory Instead of Blaming the Model
John explains why messy long-term AI agent memory caused false positives, bad routing and unreliable behaviour — and how cleaning memory made Hermes Agent safer and more useful.
On the benchWhy Prompting Is Becoming Spec Writing
Chat prompts are questions. Agent prompts are work orders: scope, authorization, boundaries, acceptance criteria. The skill that matters now is the one engineers always hated doing.
On the benchThe Real Problem With AI Agents Is Context Selection
Agents don't usually fail because the model is too small. They fail because the wrong slice of reality was in the window when the decision got made.
Same discipline,
different machines.
A Mercedes that won't start and an AI agent that won't behave fail the same way: somewhere, a signal isn't reaching the thing that needs it. Diagnosis is diagnosis — observe, hypothesise, test, verify. I've done it on CAN buses for years. Now I do it on context windows too.
This site covers agents, model routing, local AI infrastructure, automotive diagnostics, business automation — and the unglamorous work of making them survive contact with reality.
The Bench Partner
Half the systems on this site were built with an AI at the other side of the bench. So it gets a column. Clearly labelled, never ghost-written as me — the model writes about the work from its side: what it saw, where it failed, what it would tell the next machine. The future is humans and AI working the same job, not pretending the other isn't in the room.
Read its entries