Unit 01.04: Choosing the least autonomous thing that works

Work up from the simplest structure, and let a specific requirement force each step.

Five requirements, five answers

Each row names a requirement and the smallest thing that satisfies it.

The table below lists them in order.

requirement                                  smallest thing that fits
the sequence is known in advance             a script
one decision, fixed set of outcomes          a script with a rule
the next step depends on a result            an agent
it must pause for a person                   an agent with a gate
it must survive being interrupted            an agent with durable state

2 of 5 need no agent at all

Two of the five need no agent at all. A known sequence is a script; one decision with a fixed set of outcomes is a script with a rule in it.

The three that do need an agent each name something concrete: the next step depends on a result, it must pause for a person, or it must survive being interrupted. Those are requirements you can point at, not impressions.

The mistake this prevents

The mistake is starting at the bottom of the list because the problem sounds agentic. Every row you skip past adds machinery between you and the bug - a loop that can run, a permission model to get right, a state object to keep consistent.

Takeaway

Work up the list and let a named requirement force each step. If you cannot name the requirement, the previous row was sufficient.

Mark complete