GenAI Field Guide

Where it fails

Not fire and forget

For anything with consequences, one prompt is not a workflow. The reliable pattern alternates the model with deterministic checks and ends with a human.

≈2 min
1 · LLM draft Extract from theinput 2 · Code check Schema · numbers ·source 3 · LLM repair Fix what failed 4 · Code gate Run the finalchecks 5 · Human sign-off Review before release
The model drafts and repairs. Code checks what code can check. A person signs.

Rules that survive contact

Rule In practice
Anything checkable by a program is checked by a program. Dates, sums, IDs, references, formats that follow a pattern.
One job per call. "Summarize, then translate, then format as a table" is three calls with a check between each, not one prompt.
A second, independent call critiques the first. "Here is a draft and its sources; list every claim not supported by the sources." Cheap insurance, not proof.
Keep the human where the consequences are. The gate before anything goes out is a person with a name on it.

What this looks like without code

You don't need a script to use the pattern:

  1. Chat 1 drafts from the attached source.
  2. You check every number against the source table.
  3. Chat 2 (fresh) gets the draft, the source, and your list of problems, and repairs it.
  4. You read the final version and sign it, or don't.