Why a named person approves every judgment call in Orme OS
Agents are good at operational work and should not be making commitments for your company. How Orme OS separates observed facts from judgments and external actions, and what happens when a person disagrees with an agent.
On this page7
Short answer
Because a record and a forecast are only worth trusting if a person answers for every judgment in them. Orme OS lets agents log observed facts with their source, makes them propose stage, amount, close date and commitments for a person to accept, and holds every first email until a named person approves that exact message.
Why not let agents act on their own?
#An agent can read a transcript faster than anyone on the team. It cannot be accountable for what the company promised a buyer, what a deal is worth or when it will close. Those are judgments, and when one goes wrong, a person has to explain it to the buyer or to the board. So Orme OS gives agents the work and keeps the judgments with people.
There is a practical reason as well. An automated email that says something false about a buyer costs more than the time it saved, and the damage lands on the sending domain the whole team relies on.
What can an agent write to the record by itself?
#Observed facts: a meeting happened, these people attended, a message arrived on this thread. Each one is written with the agent that wrote it and the evidence it came from, and each one can be reversed.
What does an agent have to propose instead?
#Stage, amount, close date, forecast category and commitments. An agent can propose a change to any of them, but the proposal has to cite its evidence or the software refuses it. The person who owns the deal sees the proposal next to the sentence it came from, and accepts, edits or rejects it.
What needs a named person's approval?
#Anything that leaves the company. First emails are approved one message at a time, with no bulk approval. The approver sees the message exactly as it will be sent, with the full envelope, and if either changes before sending, the approval no longer holds. Approving a sequence approves its follow-up steps, and those are the only messages that go out without a fresh approval.
What happens when a person disagrees with an agent?
#The person's decision stands and is kept. If an agent later observes something that conflicts with a human correction, Orme OS does not overwrite the correction. It raises a discrepancy for someone to review, and the item stays open until it is resolved, escalating after seven days and again after fourteen.
How can you check what happened later?
#Every write and every approval goes into an audit ledger: who or what acted, what triggered it, the evidence excerpt, the change and the time. The ledger exports to CSV, and any field on the record traces back to the entry that changed it.