Enterprise / RiGi Group

Prompt Management

Treat prompts as versioned operating inputs that can be reviewed, tested, released, and rolled back.

Enterprise considerations

Make the boundary explicit.

Prompts are operating instructions. When they change, behavior changes. Prompt management means every change is reviewed, tested, released with a version, and can be rolled back, the same way you’d treat a code change.

In white water, a small change in the stroke changes the whole line. Prompts work the same way.

What to examine

Prompt Management in practice

01

Review and test changes before release.

02

Connect versions to evaluation results.

03

Roll back when results do not meet requirements.

What we look at

How RiGi reviews it

  1. 01Where prompts and agent instructions live today, and who can change them.
  2. 02Whether changes are tested against known cases before release.
  3. 03Whether you can tell which prompt version produced a given output.

What you get: A simple release process for prompts (owner, review, test set, version, rollback) sized to your team.

From our own work

An instruction meant to stop duplicate reporting (“don’t create a second system”) ended up burying the rules it sat next to. A reviewed, tested change would have caught it. Read the Agent1 case study ↗

Questions

Common questions

Is this overkill for a small team?

No. Even a shared doc with versions and a 10-case test set prevents most surprises.

Who should own prompts?

The person accountable for the workflow’s outcome, not whoever last edited it.

What about agent instructions?

Same rules. They’re prompts with more authority.

WWAIOS™, the WhiteWater AI Operating System. The governance layer that sits around any AI model or agent and checks evidence, authority and receipts before AI-supported decisions become action. RiGi Group puts it to work inside your business.