The POC Screen
Twelve questions that decide whether an agentic proof of concept can reach production — asked before you build it.
Most agentic proofs of concept are judged after the demo goes well, which is the one moment nobody is able to judge them. This screen moves the decision earlier. Twelve questions across four sections, and each question defines what earns a 0, a 1 and a 2, so two people scoring the same proof of concept land in the same place instead of trading opinions. The maximum is 24. One section is a hard gate: any zero on blast radius stops the exercise whatever the total says, because a high score elsewhere does not buy back an action you cannot undo. Three cut-lines read the result — pilot candidate, narrow it and score again, or a demo rather than a pilot. It closes with four ways to answer most of the questions before the agent exists, in ascending order of cost, each one cheaper than a pilot that fails in front of a customer.
What you can do with it
- Score a proposed agent against twelve criteria, and defend the number to someone who disagrees
- Find the rows where two people scored two points apart, which is where the team does not share a picture of the system
- Decide whether to narrow the scope of a proof of concept or stop it
- Answer most of the screen without building anything, using an evaluation set, a human behind the curtain, deliberate tool failures, or a shadow run
Published 13 September 2026