How Work Starts
Mindfront separates the work being done from the surface where people see and guide it. A job is the canonical unit of AI work. Conversations, email, CRM deals, meetings, and other business records are surfaces around that work; they are not separate execution models.
This is deliberate: people can begin in familiar places, then step closer to the AI when work needs instruction, correction, or review.
How Jobs Begin
Conversation
A person asks Mindfront to do something in chat. The resulting job remains connected to that conversation, so the person can inspect it, add context, change direction, or continue working alongside the AI.
Business Event
An integration reports that something happened: an email arrived, a meeting ended, a CRM record changed, or a Fiber service emitted an event. Mindfront’s decision engine evaluates the event and may start a job when useful work is required.
Examples include:
- A new email arrives in a monitored inbox
- A commit lands in a watched GitHub repository
- An ERP document changes: a new sales order, invoice, or customer record
- A meeting ends and its transcript arrives
- A monitored news feed publishes a relevant item
An event becomes part of the business history. It starts a job only when the decision engine or a matching standing order determines that work should follow.
Standing Order
A standing order is an explicit policy for a known event pattern. It can suppress the event, send it through judgment, or start a pre-authorized job directly. This is how repeated work moves toward higher autonomy without hiding the rule that authorized it.
Direct Command From a Surface
A person can launch work from the object already in front of them, such as a CRM deal, meeting, or other supported business record. For example, Qualify on a deal starts a job about that deal without requiring the person to restate its context in a blank prompt.
Moving Between Surfaces
The origin does not lock work into one interface. A job that began from an email or standing order can be inspected and discussed in chat. A person can launch AI work from a CRM deal, review the result, then continue manually when the situation demands it. The risk engine independently determines whether proposed external actions run automatically, receive machine review, wait for human approval, or are refused.