The overview, the approach, and the note on AI all say the same thing in different words: a person stays in the loop. This note is the detailed version of that sentence.
What the software does first
AI is used where it saves reading time:
- Pulling names, dates, amounts, and places out of long text.
- Grouping notices, laws, and news that concern a product or a route.
- Translating so a foreign-language source is not missed.
- Marking what changed since the last reading.
- Drafting a first summary for a person to check.
Those tasks are support. They are not the release.
What the person does
- Decides which question the work is actually about.
- Judges whether a source is reliable and whether a finding follows from it.
- Weighs a risk against an opportunity, and a value against a cost, with the sources in view.
- Sends the report on, or sends it back. At AetherVett, a person reviews and approves every report before it is released.
The phrase used on the site is human in the loop. It means the loop is not closed by the model. The model can propose. The person disposes.
A test you can apply
The AI note gives a simple test. If a decision would be embarrassing or costly to get wrong, a person should make it with the AI's work in front of them, not leave it to the AI. A report is built for that test. The sources are attached so the person, and later the reader, can see the work.
A chat tool can help you think the decision through. The review that releases an AetherVett report is a separate step. Both can exist. One does not stand in for the other.
What review is not
Review is not a promise that every fact in the world was found. Gaps remain gaps. Review is not legal, tax, or financial advice. Review is not a reason to skip your own reading when the decision is yours.
Review is also not a hidden score. You should be able to see the sources the reviewer saw. If a conclusion has no source line, it is not in the shape these notes describe.
Not legal, financial or other regulated advice.