Progression map
You do not graduate to a fleet. A break pushes you there.
Agent capability compounds in layers: memory, procedure, enforcement, then coordination. Skipping a layer makes the one above it unstable.
Read the original on XThe evidence card
Five levels describe a practical progression from raw prompting to an operating fleet. Each level exists because the level below it stopped carrying the work. The useful unit is not the tool; it is the infrastructure that makes the next kind of work possible.
The model
The visual is an editorial compression of the source essay: enough structure to orient a reader, without pretending that the model replaces the underlying evidence.
The argument
The site version keeps the source's operational insight while making its sequence easier to inspect, share, and connect to the rest of the archive.
Raw prompting works for isolated tasks until memory disappears. A project file helps until procedures become contextual. Skills help until quality needs enforcement. Hooks help until scope and persistence demand orchestration.
CLAUDE.md sets the boundaries. Skills carry repeatable procedures. Hooks make quality automatic. Orchestration coordinates work that is too large or too long for one context window.
Most projects do not need a fleet. The progression is useful precisely because it gives a stopping rule: build the next layer when the current one is breaking, not because a larger system sounds impressive.
The boundary
Build until something breaks; then make that break structurally harder to repeat.
The progression is a field guide, not a universal maturity score. A smaller project may be healthier when it stops at a lower level.