Vellix Business OS · Core concept
How a job actually runs, step by step.
A job is the step-by-step way your company gets something done. Run it under your written rules, and four things change.
Following up with a lead, approving an invoice, renewing a customer's proposal. The four things that change: each step is given to a person, an AI step or a tool; each step is designed to see only what that step needs; important actions stop and wait for a person's OK; and important actions are recorded — who did it, when, and under which rule. The industry name for this is a governed workflow. Ordinary automation shuffles information when something triggers it. This does the work the way your company decided it should be done.
The difference, side by side
Ordinary automation, or a job run under your rules.
| Question | Ordinary automation | Under your rules |
|---|---|---|
| What is it allowed to do? | Whatever it was wired up to do | Only what the Blueprint you approved says |
| What can it reach? | Usually one connection that can reach everything | Each step reaches only what that step needs |
| When does a person decide? | Rarely on purpose. Usually after something has gone wrong | At the points you chose, before the thing happens |
| What's the record? | A log file, if you're lucky | A name against each important action: who, what, when, and under which rule |
| What does it cost? | You find out when the bill arrives | Counted as it goes, with the budgets and alerts you set |
| How does it get better? | Someone goes in and rewires it | Changes are suggested from results someone checked, and a person approves them |
Where AI fits
AI does a step. It never runs the show.
Inside a job, some steps are done by AI: drafting a document, sorting a request into the right pile, checking an order against your rules, summing up a long email thread. Three rules hold for every one of them:
An AI step is given what it needs and expected to hand back one thing. It does not wander around the business looking for work.
Never a key to everything. What each step can reach comes straight from your Blueprint, by role and by step.
Anything sensitive is checked against your rules and, where you said so, by a person. Approving is a separate step from the AI work being approved, so an AI step is not given the ability to approve its own.
This is why we say the AI is not the product. The product is the system it works inside, and your people are still the ones deciding.
The third rule is the one people underestimate. We wrote it out at length in what a human-in-the-loop AI system actually looks like, including why approving has to be a separate step from the work being approved.
Examples
What this looks like in an ordinary business.
These are examples of how it is meant to work, not customer stories. No real customer is meant, and where the product stands today is labeled on the Standard Edition page.
A signed proposal arrives → a customer record is created in your CRM → a welcome email is drafted from your templates → the kickoff is booked against real availability → it stops: the account manager approves the whole package → everything goes out, and is written down.
Drafted from your price rules and checked against your Blueprint — then it stops, because anything over your limit needs the owner to sign off.
A new price arrives as a suggested change to your Blueprint, not as an update. Nothing changes itself; you say yes or no, and the decision is dated.
Which job would you pin down first?
Tell us about the one that eats your team's week. We'll walk through how it would run, straight, including whether it's a fit at all.