loops vs graphs, explained in plain words
same ticket, two ways to close it, and picking the wrong one burns your weekend
• loops
you set the frame, the agent picks the path
draft a reply -> check it against policy -> resolved? -> if not, loop again
use it when you don't know the path yet
• graphs
you draw the path, the agent fills every node
read the ticket -> known issue? -> fix or investigate -> draft -> QA passes? -> review passes? -> send -> log
use it when the pipeline is the same every day
how it sits on disk:
📁 support-agent
┃
┣📁 Loop
┃ ┣📁 /goal
┃ ┣📁 /policy
┃ ┗📁 /retry
┃
┣📁 Graph
┃ ┣📁 /nodes
┃ ┗📁 /edges
┃
┣📁 Company Brain
┃ ┣📁 /tickets
┃ ┗📁 /product-docs
┃
┗📁 State
┗📁 /log
both shapes read from the same company brain. skip it and the agent relearns your business on every run
start with a loop while the work is messy
freeze it into a graph once it settles
the engineer who can tell which shape a task needs is the one who gets promoted in 2026
and when one job needs both at scale, it looks like this: