# Grade your own AI setup

A one-page scorecard for anyone using AI agents to build or run a business. It maps
to the six stages in Anthropic's AI-Native SDLC playbook. Score each stage 0, 1, or
2. Add it up. The number does not matter; the empty boxes do.

- 0 = we do not do this
- 1 = we do it, but it lives in someone's head or a chat window
- 2 = it leaves a file behind that the next person or the next AI session can read

---

## Plan

- [ ] Every task starts from a written statement of who asked, the problem, the
      wanted outcome, and the constraints.
- [ ] That statement is saved somewhere a future session can read it, not just
      spoken once in a chat.

Score: ___ / 2

## Design

- [ ] Before building, the AI is handed the rules it must follow: your voice, your
      brand, what you never say, any compliance line.
- [ ] Those rules are a required step, not a thing you remember to paste in when
      you think of it.

Score: ___ / 2

## Build

- [ ] The AI works from a plan, not a vibe.
- [ ] The plan says how you will prove the work actually worked before you trust it.

Score: ___ / 2

## Test

- [ ] Something other than the AI that wrote the work checks the work: a test, a
      second AI, or a person.
- [ ] A failed check actually stops the work from shipping. It is a gate, not a
      suggestion.

Score: ___ / 2

## Deploy

- [ ] Nothing reaches your customers without a preview step you can look at first.
- [ ] The risky changes (money, customer data, anything you cannot undo) need a
      human to sign off.

Score: ___ / 2

## Maintain

- [ ] Something watches for problems and writes them down without you asking.
- [ ] When it finds one, it records the likely fix and who owns it, not just "this
      is broken."

Score: ___ / 2

---

## How to read your score

Do not chase a perfect 12. Look at the zeros and ones.

- The cheapest win is almost always in **Plan**: write the intent to a file before
  you start. One paragraph. The next session reads it instead of losing the thread.
- The most expensive gap to leave open is in **Test and Deploy**: if the AI that
  writes the work also approves it, you will ship its mistakes at the speed it makes
  them.
- The gap most people forget is in **Maintain**: a tool that files "this is broken"
  is a to-do list. A tool that files "here is the problem, the likely fix, and the
  owner" is the start of the thing fixing itself.

Made by Bluhook. We build and run AI systems for businesses, and we tell you what we
learn while we do it. bluhook.com/learn