Field notes

Practical notes for teams shipping AI.

Architecture choices, routing patterns, model economics, and local workflows for building an accountable AI platform.

A routing decision selecting a teal model path among glass model nodes
route / score / select
Routing2026-07-105 min read

Model routing that explains itself

Why capability, cost, latency, reliability, and organization preference belong in one visible decision.

RoutingPolicyCost control
Read note
A glass repository passing through two quality checkpoints
review / test / ship
Developer workflow2026-07-084 min read

Pre-commit quality gates for AI-assisted engineering

Use fast local checks before a commit and keep the expensive confidence checks at pre-push.

Git hooksReviewDeveloper workflow
Read note
A luminous request signal splitting across model lanes and selecting an efficient route
mix / measure / save
Cost engineering2026-07-128 min read

The model combinations that lower cost without lowering the bar

A role-based routing playbook for search, implementation, review, and architecture work—with the assumptions behind every savings estimate.

Model mixSavingsBenchmarks
Read note
A glass repository passing through managed quality gates
tools / skills / sync
Control plane2026-07-127 min read

The company controls that reach every developer binary

How managed policies, MCP servers, skills, instructions, and Git hooks move from the dashboard to a local, inspectable runtime.

MCPSkillsManaged config
Read note
A routing decision selecting a teal model path among glass model nodes
trace / explain / improve
Observability2026-07-126 min read

Trace an AI request from prompt to provider receipt

A field guide to the policy decision, model score, fallback chain, tokens, latency, cost, and audit event behind one request.

TraceAuditDebugging
Read note