Journal

Field notes from sovereign builds.

What we learn shipping agentic systems onto hardware our clients own. Written plainly, illustrated properly, no hype.

  1. № 004 13 min read

    AI sovereignty is a spectrum.

    A practical guide to where your AI runs, who can operate it, and who owns what it learns, with the UAE as a concrete example.

    Read the essay
  2. № 003 14 min read

    The secret sauce to building specialized AI agents.

    There are two ways to make a model an expert in your domain: fine-tune it, or hand it the right knowledge at the right moment. This is the second way, one failure at a time.

    Read the essay
  3. № 002 8 min read

    Your model shouldn’t see your entire toolbox.

    Eighty tools is prompt bloat wearing a new coat. Keep the top level boring, and let the rest of the toolbox wait for its cue.

    Read the essay
  4. № 001 7 min read

    Your agent doesn’t need a better prompt.

    It needs a thicker harness. Why patching the system prompt never converges — and what to build instead.

    Read the essay

Prefer a feed? Subscribe via RSS