01
Trigger
What starts the run — schedule, webhook, or human button?
Lesson 06 / 07
Scan → Monitor · ~10 min
Agents and workflows
Sketch a real agent: inputs, tools, human gate, output — the UP-ON operating shape.
An agent is a path with permissions
Useful agents are boring on purpose: clear trigger, allowed tools, explicit stop points for humans, and an output someone owns.
Scan → Diagnose → Design → Deploy → Monitor is the same idea at company scale.
01
Inputs
02
Tools
03
Human gate
04
Output
Fig. 06 — Four slots every agent needs
Design questions
02
Permissions
What may it read? What may it write? What is forbidden?
03
Gate
Where must a human approve before anything external happens?
04
Monitor
How will you see silent failures before customers do?
Exercise
Sketch one agent
Fill the four slots for a workstream you care about (sales follow-up, reporting, proposals…). You are designing the system — not chatting.
0 / 4 slots filled — stay for your notes; progress is local to this browser.
Check
What is the most important missing piece in “an agent that emails customers automatically”?