Architecture playground
Build a graph, then break it
Start from one of seven graphs or an empty canvas. Add agents, tools, checks, reviewers, checkpoints and safeguards; connect them with flows, branches, loops and failure paths; set what each branch takes and what each person answers; fail a node and replay. The run is a learning simulation in your browser: the same engine as the pattern pages, no model calls, nothing leaves the page.
Press Run to execute the graph as drawn. Change a branch, a reviewer’s answer or a failure below and run again: the trace changes because the graph does.
Shared state fields
requestriskplanattemptsdecision
Select a node to rename it, change its kind, move it, connect it or remove it. Add nodes with the buttons above.
Things to try
- Make the tech lead reject. The plan loops; reject again and again and the step budget ends the run. A real graph needs a limit on that loop too.
- Fail Open PR with retry off. Then add a failure edge from it to the tech lead and run again: recovery becomes a designed path.
- Remove the checkpoint and ask what would happen if the tech lead answered tomorrow.
- Load Security investigation and look for the only node that can change anything.
The Agents & Humans Briefing
How humans and agents work together: the day's field notes and the patterns behind them, in one short email around noon UTC.
Free. No spam. Unsubscribe anytime.