Software Engineering

An office-envelope analogy for keeping AI planning, policy authorization, execution, and audit evidence separate.

BMAD and Micrantha converge on several autonomous-development mechanics, but differ in where context, identity, authority, durable state, repository truth, and evidence live.

The AI-native SDLC gets the bottleneck shift right. My own architecture pushes the idea further: generated artifacts are cheap, but trusted acceptance and authorized effects must remain explicit.

What a decade of self-hosting taught me about reliability, portability, Canadian infrastructure, and AI

A reproducible walkthrough of the Anthesis Governance Lab: 7 canonical governance scenarios, 27 broader governed-action scenarios, and 24 inference-integrity cases over recorded evidence.

How Dubnium is evolving from local model serving into a governed agentic control plane with a supervisor gateway, bounded memory, specialist routing, and deterministic execution boundaries.

How Dubnium uses declarative schedules, systemd timers, workflow dispatch, and a separate agentic control plane without turning the scheduler into a hidden workflow engine.

Dubnium is a NixOS-based personal super server built to stay legible: safe defaults, explicit modes, layered mutability, local AI boundaries, and an optional lab surface.

A practical architecture for a prompt-driven, spreadsheet-backed CareerOps workflow: discover and rank roles, curate evidence, tailor resumes through staged prompts, and preserve human accountability while preparing for bounded Dubnium automation.