← Back home

Writing

Long-form notes on architecture, AI governance, and shipped project work.

Verification Criteria That Derive From Requirements (Not Invented Next to Them)

8 min

Change a rule on an Event Modeling board and watch a previously-passing build fail — with zero code changes. The one-way path from a spec artefact to acceptance criteria, an oracle gate, and drift detection.

event-modelingspec-driven-developmentverificationagentic-ai

Pi vs. Antigravity vs. Orca: One Job Per Harness, Not One Harness to Rule Them

7 min

Why I retired Orca, kept Antigravity for exactly one job, and picked Pi as the headless runner underneath an oracle-gated build pipeline — and the measured result: Pi came in 27–55× cheaper than Claude Code per successful build on the same tasks.

agentic-aideveloper-toolsmulti-agent-orchestration

When Your Governance Layer Is the Thing That's Wrong

6 min

An autonomous coding agent solved a HIPAA-shaped PHI anonymizer correctly on its first live attempt — and a deterministic harness failed it five times anyway, because the wrong file was under scrutiny.

agentic-aigoogle-adkgovernancecase-study

The Orca Agent Development Environment: A Practitioner's Guide

6 min

A practitioner's field guide to Orca, an open-source multi-agent development environment — what the fan-out pattern actually buys you, and where git worktree isolation quietly breaks down.

agentic-aideveloper-toolsmulti-agent-orchestration

Cybersecurity Assessment: Gunra Ransomware (AA26-222A)

3 min

A practical breakdown of CISA advisory AA26-222A on the Gunra ransomware-as-a-service operation — attack vectors, and prioritized mitigations for edge devices, identity, and lateral movement.

cybersecurityransomwareincident-response

Converting Event Modelling Board to GitHub Specs

2 min

A six-step pipeline for turning an Event Modeling board into GitHub Spec-Kit specs, keeping the architectural constitution separate from fast-changing feature slices.

event-modelingspec-driven-developmentautomation