How the score works
Each task gets points for five things. How many hours a week it takes, how regular the input is, whether the rules are written down, whether there are past examples to test against, and how easy a mistake is to catch. The most a task can score is 14.
The questions are the ones that decide whether a project works. Volume makes it worth doing, regular input and written rules make it possible, past examples make it testable, and the cost of a mistake decides how much a person stays involved.
What the verdicts mean
A strong first candidate is frequent, rule-based work with examples to test against. Start it in shadow mode, beside the team, before it acts on anything.
Where the work is mostly judgement, the right design is decision support. The agent gathers and ranks, and a person decides.
Where a task takes under two hours a week, an agent is the wrong tool. A form, a template or a script will fix it for less. Where there are no past examples, collect them first, because without real cases there is no honest way to test what you build.
What it leaves out
It is a first sort, not a design. It does not know your systems, how clean your data is or how the team feels about the work, and each of those can change the order. Use it to choose what to look at properly.
Nothing you type leaves your browser unless you press send.