LoopArchitecture

Agentic Loop Storming

A short, collaborative workshop for discovering the agentic loops between your systems, in the spirit of Event Storming. Map the systems, connect them into loops, name them.

What you need

A wall or a virtual canvas, the team (people who feel the pain, own the systems, do the work), and sticky notes in three colors:

system
A place a loop reads from or writes to, a repo, a store, a tool.
loop
An agent that connects systems: it uses some and writes back to others.
trigger
What turns a loop, a schedule 🕐, an event ⚡, or manual.

The steps

  1. Define the systems. Focus on data, source code, and storage. One sticky each; note the URL or repo.
  2. Brainstorm the loops. Which agentic loops make sense between these systems? Look for the manual, between-systems work.
  3. Draw them as edges & name them. Arrows for uses and writes back; a verb-first name and a one-line prompt each.
  4. Add the trigger icons. Mark when each loop turns with an emoji (below). A loop can have more than one.
  5. Prioritize. Impact × confidence. Start top-right.
🕐 schedule
Runs on a cadence, a cron like 0 6 * * *.
⚡ event
Runs on an event, e.g. pull_request.merged, push.
📅 once
Runs a single time, an ISO timestamp.
👆 manual
Runs on demand.

Example: a pen & paper wall

A real Loop Storming session, the systems and loop edges that became this site's example architecture, drawn by hand:

A hand-drawn Loop Storming wall: systems (CI/CD, Logs/Traces/Metrics, Docs, Jira, Confluence, App/Azure, Terraform, ISO 27001, OneDrive, Slack) connected by loop edges.

From wall to architecture

Refine the wall into one Loop Architecture file, then let looparch validate, draw, and publish it.

$ looparch init your-org
$ looparch view your-org.looparch.yaml      # all loops in one interactive diagram
$ looparch publish your-org.looparch.yaml   # → Claude Code routines