Latest · Technology & Ventures · 3 min
The fallback that works
The failure that cost me the most this year never broke anything. A backup path kept producing fresh output, every check read green, and the real source had been dead for twelve weeks.
Stephen Oldfield · Chief AI & Transformation Officer, Research Results
That is how I ran a public agency for nearly six years, and how I run a research firm's work with AI now. Everything on this site was built the same way, on my own time, from the same desk.
Plate 01 One brain, four arenas, every build All builds →

Plate 02 / From the Ledger
Latest · Technology & Ventures · 3 min
The failure that cost me the most this year never broke anything. A backup path kept producing fresh output, every check read green, and the real source had been dead for twelve weeks.
The truck took an afternoon because the tractor didn't. What I built the first time wasn't a tractor log — it was a way of keeping any machine honest.
Most AI systems are judged by what they can do. The ones worth trusting are defined by what they were built to refuse — and the refusal list is where the people stay.
Plate 03 / Selected work
What I built, the choices behind it, and what the evidence can support.
Fig. 03.1 — Stylized map · anonymous graph data

Information has a route in, a place to live, and a human review before it becomes a decision.
Fig. 03.2 — Reconstruction · all figures invented

A number needs a source, a date and a boundary. The view should show what is still unknown.
Fig. 03.3 — Actual software capture · personal project

The audit found the missing records. The cockpit made them harder to forget.

Plate 04 / Behind the work
My day job is helping a research firm work differently because of AI — which tools, which workflows, and which decisions stay with people. I spent the fourteen years before it inside operations, and that is still how I see the problem.
The rest of what's here I build on my own time. I'm also a husband and father of two boys. Some of what I build belongs in a board meeting. Some of it belongs in the barn.
I'm glad to compare notes on AI in a working organization, on operating instruments, or on any of the builds.
Start a conversation →