Field Notes

Practical writing from the coordination layer, not framework hype.

The Mycel blog is where we publish the arguments, failure modes, and design tradeoffs that emerged while building protocol-first text systems and operating multiple AI coding agents in one repository.

The goal is not to announce "another agent framework." The goal is to explain what actually has to work when repository state, Git boundaries, and handoffs must survive interruption and parallel work.

Publishing Model

Markdown in the repo stays the source of truth. These pages are static presentation layers.

Current Focus

Early posts center on coordination primitives: registry, workcycle, handoff quality, and Git ownership.

Language Scope

v1 publishes English first. Localized blog pages can arrive later for durable, strategic posts.

Latest

Three field notes are live now

The series now has a framing post, a workflow post, and a metadata post. The newest article explains why token usage summaries and commit trailers belong inside the coordination layer rather than being treated as optional logging detail.

Agent Coordination Field Notes Part 3 7 min read

Why Token Usage and Commit Trailers Belong in Agent Workflow

Why usage summaries, commit trailers, and repo-native tooling help both humans and agents understand cost, ownership, model context, and landed boundaries.

Published 2026-04-17 token-usage git metadata
Agent Coordination Field Notes Part 2 7 min read

What Actually Manages Agent Workflow: The Repo-Native Scripts We Use

A concrete tour of the registry, workcycle, mailbox, commit, push, checklist, and planning-refresh scripts that keep our coordination layer visible and durable.

Published 2026-04-10 workflow scripts registry
Agent Coordination Field Notes Part 1 6 min read

Multi-Agent Coding Is a Coordination Problem, Not Just an Orchestration Problem

Why real-world multi-agent repo work breaks down on ownership, lifecycle, and continuation surfaces long before it runs out of clever delegation strategies.

Published 2026-04-03 multi-agent handoff git
"For real teams, multi-agent coding is a coordination problem before it is an orchestration problem."

Series

Agent Coordination Field Notes

This first series turns repo-local practice into public writing. It is intentionally technical, concrete, and explicit about overhead rather than treating process as invisible magic.

  1. Multi-Agent Coding Is a Coordination Problem, Not Just an Orchestration Problem
  2. What Actually Manages Agent Workflow: The Repo-Native Scripts We Use
  3. Why Token Usage and Commit Trailers Belong in Agent Workflow
  4. Why Chat Memory Is Not a Coordination System
  5. The Smallest Useful Primitive Is a Registry, Not a Swarm