APPLICATION

Javier Ponz, writing from Aberdeen. For ten years at Apple Retail I worked where a technical system meets somebody who depends on it and cannot repair it themselves. I run open agent systems now, and the interesting problems sit on the same seam.

I am not a research engineer. What I have is operational: I run these systems every day, I break them fairly often, and I can usually say precisely which assumption failed. That competence is easy to undervalue from inside a lab, where progress is legible as a benchmark. Adoption is settled in the half hour after something breaks, when the user decides whether to open it again.

One conviction has held through every role I have had, which is that leading is serving. It meant the same thing at a repair bench, teaching somebody else to work that bench, and running a shop floor: the only honest measure is whether the person you are responsible for leaves with more capacity than they arrived with. Software gets the same test. Does the person end up more capable, or only more reliant.

What I would like to do at Nous is technical support engineering: the work where somebody has to be technical, patient and a competent writer at once, and where a user's broken situation becomes either a fix or a clear account of why it is not one. Ten years of exactly that at Apple is the part of my background that transfers without translation.

Further out, the work I want is forward-deployed engineering. I am not ready for it today and would rather name it than have it inferred. If Nous is somewhere that lets a person carry real support load while learning on the job, that is the path I would like to be on.

↓ Download CV (PDF)

An operating system malleable enough for an agent to inhabit strikes me as a more interesting problem than a larger model, and probably a nearer one. Most personal computers are appliances: quick, sealed, arranged according to decisions somebody else made years ago and cannot easily revisit. For a long time that trade was reasonable, since the alternative was a machine you maintained rather than used.

Agents change what the trade costs. An agent's usefulness is bounded by the surface it can reach, and on a sealed appliance that surface is conversation and very little else.

Two efforts seem to me to be converging on this from opposite ends. Hermes and MCP decompose capability into things you can inspect: a skill is a Markdown file, a tool is a server you can read. Omarchy approaches from the system side, treating the desktop as configuration its owner is expected to edit, and describing itself as an OS for the age of agents. Between them sits the machine I would like to use, one that is proactive and shapeable by the person living in it.

You cannot build a malleable machine on a stack you are forbidden to read, and whoever holds the primitives sets the ceiling on how far the agent may help. Open weights and open primitives are what keep that ceiling in the owner's hands, which is the practical reason I care about the position Nous has taken, beyond agreeing with it.