May 18, 2026 · 4 min read
We exist because most AI agency work is theatre
Demos that wow leadership and break in production are the default — and that's the gap we built ncodelab to fill.
Blog
Notes from the workshop.
What we're learning building agents in production. Less 'the future of AI' takes, more concrete tooling, evals, and failure modes.
May 12, 2026 · 6 min read
Write the eval before you write the prompt
A practical pattern for building agents you can trust — start with a scoring rubric, end with a measurable score, and you'll never ship a "vibes-based" system again.