AI workflow checklist

Real work is a chain of steps, not one prompt. These checks help you design the chain on purpose. The main question is where a person reviews the work, and why there.

The checklist

1. 1

   Find the task you repeat

   Look for AI work you do every week and rebuild from memory each time.

2. 2

   Save what worked

   Keep the request, the inputs it needs, and one good output together in one place.

3. 3

   Split the task where you need to check

   Break the work into steps only where an intermediate output needs review.

4. 4

   Define each step’s input and output

   Write what goes into each step and what a correct result from it looks like.

5. 5

   Place review after the riskiest step

   Put the check where an error would spread into everything after it.

6. 6

   Match the tool to the step

   Use a purpose-built tool for structured jobs and a stronger model where reasoning matters.

7. 7

   Write it down for someone else

   Record the steps, requests, formats, and review checklist so a colleague can run it.

8. 8

   Run a pre-mortem

   Assume a bad output shipped. Name the step that failed and add a check there.

Red flags

• You review only the final output, after every step has built on the first one.
• The workflow has five steps and no step has a defined output.
• Only you can run the workflow, because it lives in your memory.
• Every step uses the same tool because that is the one you had open.

At each band

E1 · Foundational

Saves a request that works and reuses it instead of rewriting from memory.

E2 · Proficient

Splits work into steps where an intermediate output needs checking, and only there.

E3 · Distinguished

Chooses a tool or model class for each step and documents the workflow for others.

E4 · Exceptional

Places review exactly where errors compound and can explain the reason for each checkpoint.

Summarized from the Emeri study guide Putting it together, free at emeri.org/learn/putting-it-together.