About

I’m Cameron Jones. I build reliable infrastructure around LLMs and coding agents.

My background spans 25 years across infrastructure, SRE, cloud, and enterprise architecture. That experience shapes how I approach AI systems: production behavior matters more than demonstrations, and autonomy is useful only when its authority, failure modes, and operating boundaries are explicit.

My current work focuses on inference routing, typed agent runtimes, evaluation, human-approved automation, and observable control planes. Across these projects, I favor deterministic offline tests, explicit budgets, structured contracts, minimal retention of sensitive content, and honest statements about what the available evidence does—and does not—establish.

A recurring principle is deterministic core, sparse LLM: code owns facts, state, policy, authority, and validation. Models contribute interpretation and planning where their judgment is useful.

Some of the projects here are active systems. Others are concluded experiments whose original premise did not survive testing. I keep the negative results public because revising a design when the evidence changes is part of the work.

You can find me on GitHub, X, and LinkedIn, or email me at hello@cameronqj.com.