What do the AI agents in Omni actually do?
They do the analysis work on your model. Each agent is specialised rather than general: one reads sources and structures what it finds, another answers questions about how the work runs, another proposes changes to a process. Proposals are reviewed before they are applied, so an agent never quietly rewrites your model.
How AI agents works in Omni
Specialised, not one general assistant
An agent that structures source material is doing a different job from one that argues about a future state. Separating them is what keeps the answers specific.
Grounded in your model
Agents answer from the model and the material behind it. When the model does not contain the answer, that is what you get told.
Proposals you approve
Changes arrive as suggestions against the current model, accepted row by row, with what they would alter made explicit and what the agent could not do shown with the reason. Nothing is applied until someone accepts it.
It reads what you are looking at
Ask a question while a record is open and the agent already has that record in view, so the question can be about this one rather than a description of it.
The conversation is a working session, and it survives
A thread has a name, belongs to the engagement it was about, and is still there tomorrow. Thinking out loud costs nothing, because the record only changes at the moment somebody accepts a proposal.
What you get
- Answers in plain language that name the records they were drawn from
- Analysis that keeps pace with the business rather than a quarterly review
- An AI that names what it does not know instead of inventing it
Common questions
How do you stop the AI making things up?
Nine enforced rules, described in full on how we constrain the AI. The short version: answers are grounded in your material, unknowns are reported as unknowns, and figures are never invented.