Choose the least autonomous system that works.

Six visible criteria. Four possible approaches. No model call and no assessment answers leave this page.

Input panel

Task profile

0/6

01AmbiguityHow fixed is the path from input to output?

Clear rulesSome interpretationDiscovery is part of the task

02RepetitionHow often will this task run?

Once or rarelyRegularlyContinuously or at volume

03Objective verificationCan correctness be checked mechanically?

Mostly subjectivePartlyYes, with clear checks

04Error costWhat happens when the task is wrong?

Easy to undoNoticeable reworkMaterial or irreversible harm

05Required contextHow much context must be combined?

A few known inputsSeveral connected sourcesBroad, changing context

06Human judgmentHow much judgment does a good result require?

Almost noneJudgment at exceptionsJudgment throughout

Decision rule: ambiguity + context + judgment + low objective verifiability introduces agent assistance. High error cost adds governance. Otherwise, prefer explicit automation.

email twitter linkedin github