METHOD2026-06-12 · 8 minPLANNED
Why I write acceptance specs before I prompt
The single highest-leverage habit in AI-assisted delivery — with the rework data behind it.
07 · WRITING
SAMPLEplanned titlesLong-form, on building systems with AI
Fewer, denser essays. Each one is grounded in the lab — methods that earned their place across real delivery cycles.
EVALUATION2026-05-29 · 11 minPLANNED
Hallucinations are a validation problem, not a model problem
Stop waiting for the model to stop lying. Build the gate that catches it and measure what it catches.
SYSTEMS2026-05-10 · 14 minPLANNED
A field manual for shipping AI-assisted systems in public
The full lifecycle, the gates, and the evidence trail — written as a runbook you can actually follow.
PRACTICE2026-04-22 · 9 minPLANNED
Confidence is not calibration: reading model self-reports
What self-reported confidence does and does not tell you, with the correlation data.
Essays publish here as they are finished. Until then, the working notes live in the Build Logs.