Martin | Now

A full example

Every morning each support owner gets a short report on their escalations: what is new, what is stuck, and what looks worrying. Nobody writes it. Agents do.

Full pipeline graph: schedule, list reportees, report agent with filter, search, label and account tools, an inspector sub-agent, and a store step, with its input and output tables shown alongside

How it fits together:

  • At 8:00 the pipeline starts and splits into one branch per support owner, all running at once.
  • In each branch a report agent looks through that owner's escalations with a few tools: filter by owner, see how labels are distributed, search similar cases, look up the account.
  • When a case looks concerning, it calls an inspector agent that reads the full history of that case and reports back in a few lines, so the report agent stays focused.
  • The finished report is saved to a table and shown to the team.

The notes on the canvas are part of the pipeline, so whoever opens it next, a person or Chip, knows why it is built this way.

Pipeline canvas, first part of the graph

Pipeline canvas, agent and tool nodes

Pipeline canvas, storage branch

And a run of it, where every step's input and output can be inspected:

A finished run with every node green and its per-node outputs

It also has tests: a pretend list of owners goes in, nothing is saved for real, and a check looks at what the agent wrote. That is the test shown on the tests page.