Field notes as a publication layer

Agentic Design School publishes two kinds of writing today, and there is a gap between them. The articles are long-form lessons: each one starts as a research run, gets a visual plan, goes through compliance notes, and ships with diagrams, tables, and sources. They are written to stay correct for months. On the other side sits Applied AI Signals, the personal blog, where short dispatches go out almost daily: a tool release, a workflow observation, an operator's read on what is worth testing this week.

The school has been missing the middle. Most of what we actually learn comes from small experiments inside this repository — a sync script that works, an MCP endpoint that refuses to authenticate, a prompt structure that cuts a review loop in half. Those findings are too small to become articles and too specific to the school to belong on the personal blog. Field notes are where they go.

Every field note follows the same five-part spine, adapted from the editorial method on Applied AI Signals and tightened for design work. The structure is the promise to the reader: a note is short, but it is never vague.

The spine also makes notes cheap to write. Because each section has one job, a note can be drafted from an experiment's leftover artifacts — the status JSON, the screenshots, the commands in the terminal history — in under an hour, while the details are still accurate.

• Signal — the observation, release, or experiment that triggered the note, dated and sourced.
• What we tried — the actual setup, with commands, file paths, and constraints, not a sanitized retelling.
• What changed — outcomes stated plainly, including the parts that failed or stayed blocked.
• Evidence — artifacts a reader can inspect: generated files, snapshots, payloads, diffs.
• Next test — one concrete thing to test this week, because a note that ends in conclusions instead of a test is just an opinion.

Notes come in four formats. Tool experiments report what changed when an agent was given real design constraints. Workflow notes capture reusable operating loops for briefing, critique, revision, and QA. Case studies show before-and-after examples from product work, books, and prototypes. Signal notes adapt an external dispatch — usually something first captured on Applied AI Signals — when it is directly useful to designers working with agents.

Whatever the format, the standard is the same: claims are dated, blockers are reported as content rather than hidden, and anything that cannot be backed by an artifact in the repository or a cited source does not ship. When a note outgrows the format — when the experiment turns into a workflow other teams can adopt — it graduates into a full article and links back to the note that started it.

The launch batch comes from work already done. Two tool experiments document the canvas-sync project: pushing this site's design system and page snapshots into OpenPencil, and the partly blocked, partly successful attempt to do the same into Figma over MCP. Two signal notes re-ground recent dispatches for designers: HTML artifacts as design deliverables, and extracting design systems from live websites. Each one ends with a test you can run on your own project this week.

The cadence target is one to two notes per week, distributed through the existing newsletter path on this page. If a note saves you one wrong turn in your own agentic design setup, it has done its job.