Travis Shea

I lead complex technology programs and independently research how AI systems fail, how architecture contains those failures, and what it takes to operate capable AI on infrastructure you control.

Technically Acceptable is where I publish that work.

The subject matter spans AI agent security, private and self-hosted AI, systems architecture, and the operational tradeoffs underneath them.

The writing comes in two forms.

Essays are structured, evidence-backed mini reports. They are where experiments, arguments, technical judgments, and larger ideas get worked through properly.

Notes are the live layer between them: short observations, links, intermediate findings, reactions, and provisional judgments that are worth publishing without turning them into a report.

Some of the writing comes from longer-running technical Programs.

Sentinel is the first. It is an experimental program testing what happens when AI agents encounter hostile content and whether system architecture can prevent model failure from becoming privileged action.

It will not be the last Program.

My background is in leading and operating complex technical work across defense technology, test and evaluation, systems integration, field operations, and program execution. I use that operating perspective here, but Technically Acceptable is independent work. Public writing uses public sources, synthetic material, and my own experiments. It does not use non-public employer or customer information.

The standard is simple: show the work, separate observation from inference, keep claims smaller than the evidence, and say what a result does not prove.

Contact

Contact: [email protected]