Field notes.
On agentic engineering operating models, governance, and what software delivery becomes when humans and agents build together.
Cost-Aware Model Selection: When Local Beats Frontier
The most capable model is not always the correct choice. A routing policy that defaults bounded tasks to a local engine, with a CLI you can run yourself, and two real bugs found while proving the policy worked as documented.
Governance That Survives the Agent's Own Confidence
A rule that only applies when the agent recognizes it's in a risky situation isn't governance, it's the agent's own judgement wearing a policy label. Four incidents from three different systems, and what real governance looks like instead.
Context Is the Architecture
The code is not what a future agent or engineer resumes from. The decision log is. Why context needs the same design rigor as the services it explains, and a CLI that audits whether yours is holding up.
LLM-as-Judge, and How Not to Fool Yourself
A model judging another model's output sounds confident by default. Confident is not the same as calibrated. The measurement methodology that tells the difference, and a bug the methodology caught in its own tool.
The Struggle Is the Point: What Agentic Harnesses Do to How We Learn
A rubric fix that stops a human from spotting the answer early does nothing against a coding agent that already knows it. What a Rust workshop learned about the difference, and the honest, bounded fix it shipped instead.
The roles don't disappear. They move.
Agentic delivery doesn't remove the reviewer, the architect, or the PM from the picture. It moves their judgment one level up the stack, from doing the work to defining and gating the workflow that does it.
Tools are easy. Operating models are hard.
Every vendor is selling you the model. Nobody is selling you the plumbing that makes the model's output something you can still trust six weeks later. That plumbing is the actual hard part, and it is the whole reason this factory exists.
How we built this site with Hekton
The lazy way to build this site was to ask a chatbot and paste the result. Here is what happened when I ran it through an agentic engineering factory instead: the gremlins that did the work, and why the output was cheaper to make and safe to trust.