Lesson 01 · 5 min 5 min

The rules the AI follows

What it does, what it does not, and what happens when it is off.

Galilei's AI shows up in several different places, and in every one of them it follows the same rules. They are worth knowing before the features, because they are what tells you how much to trust what you are reading.

Deterministic first, AI second

This is the rule you notice most. Everything that can be computed — temperatures, counts, dates, scores, metrics — is computed in the database, with no model. The AI only adds narrative or recommendation on top of those numbers.

The practical consequence is large: no figure you see on the platform was invented by a model. The day plan picks among candidates a rule computed, the evaluation scores against metrics that arrived pre-aggregated, and the insights rest on the client's real browsing.

Human in the loop

The AI proposes; you review and send. It is mandatory on money, on legal matters, and on any communication going out to a client. You will see the sentence written on the screens themselves: «Review and edit before sending. AI only suggests; you send».

Language by recipient

Not by whoever is writing. A draft for a client who speaks English goes out in English even if you work in Spanish; the interface and your digest emails come in your language. That is lesson 05.

What happens when there is no AI

The platform has to keep working without it, and it does. If your brokerage did not turn it on, if your plan does not include it, or if the monthly quota ran out, the screen tells you plainly instead of failing in silence.

  • The Pulse still shows your KPIs, your level and your tasks, just without a written day plan.
  • Signature field suggestions still come out, computed from the document geometry.
  • You lose the narrative, not the dashboard.

Where the spend is visible

Every call to the model is logged with its feature, its tokens and its cost. The broker sees it under My Agency → AI usage, per person and per feature. There is no hidden counter and no surprise invoice.

Tip: If a suggestion looks odd, the useful question is not "did the AI get it wrong?" but "what data did it see?". The answer is almost always in the file: a lead with no recorded activity produces generic suggestions, because there is nothing else to work from.