Travis Shea

I investigate how AI systems work, where they are useful, and what it takes to operate them reliably. My work combines independent experiments, primary-source research, and analysis of the technical and operational choices behind adoption.

Technically Acceptable is where I publish that work.

The focus is AI, defense technology, and security, with particular attention to AI agents, self-hosted systems, systems architecture, and the tradeoffs involved in choosing, integrating, and maintaining these technologies.

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.

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: travisqshea@gmail.com