Choose your operating model
Before you hand work to an agent, decide what it may change.
Start with the lightest operating model that fits the work. The more an agent can affect customers, revenue or operations, the more explicit its limits should be.
See a proposed package before sharing contact details. A person then reviews your answers and sends a scope and figure.
Three operating models
Match the control model to the work.
- For repeatable, low-risk tasks
A fixed workflow with an AI step
The path is defined in advance, with AI used for a bounded step such as classification, drafting or routing.
What it is good at
Choose this for work that is genuinely predictable: intake, routing, reminders and other tasks where the next action is already known.
Where it strains
It becomes harder to manage when the work requires judgment, moves across systems, or can create a customer, financial or operational consequence.
- For agents you already operate
A guard bolted on in front of your agents
You keep your existing agents and add a separate layer to inspect or constrain their requests.
What it is good at
Choose this when you need to introduce controls around agents that already exist and cannot be rebuilt immediately.
Where it strains
You still need a clear answer for who owns authority, approvals, evidence and recovery when an action crosses into a consequential system.
- For work that needs accountable control
A governed operating layer
VertixOS connects the work, assigned access, rules, approvals and evidence as one operating model.
What it is good at
Choose this when agents affect customers, revenue, delivery or internal operations and your team needs to define the limits before work begins.
Where it strains
We are early. What we lack is listed under when to choose someone else.
Five questions
Ask any agent vendor these five.
Put them to everyone on your shortlist, in these words, and put them to us. Our answers are underneath so you can hold us to them.
- 01
“What may this agent access and change?”
The useful answer names the tools, data and actions in scope—not a vague promise that the agent is secure.
Partly live. The risk tier of each action decides what happens next, and irreversible actions wait for a person. Separate identities for each of your agents are in preview.
- 02
“Which actions require a person to approve them?”
The most important boundary is not whether a model is confident. It is whether a consequential action can proceed alone.
Preview. Irreversible actions wait for a person. VertixOS separates the person who requests or prepares work from the person who approves it; an agent cannot approve its own action.
- 03
“How do you distinguish completed work from correct work?”
“Completed” can mean the process reached the end, not that the result was right.
Live for software delivery. A run does not close without evidence and two approvals, one from a reviewer that did not do the work. For replies, documents and record changes, that check is being productised.
- 04
“What evidence will my team see after the work runs?”
A usable operating model makes it possible to review what happened, why it was allowed and where a person intervened.
Live for governed software delivery. For governed software-delivery runs, work does not close without evidence and independent review. Equivalent customer-facing evidence paths are being productised and should be scoped before purchase.
- 05
“What is the smallest safe first use case?”
A credible rollout starts with one meaningful workflow, clear limits and a way to learn before expanding the agent's authority.
No price list. We scope one workflow first. Five quick questions show a proposed package; a person then reviews the boundaries and sends a scoped figure. There is no self-serve workspace today.
Not a fit?
When to choose someone else.
You need a self-serve workspace this quarter
VertixOS is scoped and built with you today. If you need a card-on-file sign-up and instant provisioning, start with a product that already offers it.
You need commissions, carrier contracting and agent portals now
Those back offices are years of detail. We do not have them and will not have them this quarter.
You want to sign up with a card this afternoon
There is no public sign-up or billing yet. Our work is scoped with you, then quoted.
You need single sign-on and user provisioning on day one
Both are designed, not built.
You need document AI at volume
Classifying and extracting documents at scale is a product of its own, and not ours.
You only want a chatbot
The safety layer is the costly part of what we do. Skip it if nothing you automate can cause harm.
When we are the right answer.
- Your security lead asks “what stops it?” The answer sits in the system itself, and the risk table is published so you can inspect it before you buy.
- You want to keep your choice of model. Several model families sit behind one set of rules, with failover, and the model is chosen per workspace (in preview). Bring-your-own-keys is on the roadmap.
Questions
Two more questions.
Tell us what you would hand to an agent.
Five quick questions. A person replies with a scope and figure, or tells you it is not a fit.
