Seven questions for a company that builds AI agents on Priority
The questions work for judging any vendor, us included. The last column says how we do it.
| Check | Ask the vendor | How we do it |
|---|---|---|
| Read and write | Does the agent only answer, or also record in Priority? | Both. Every write is shown for approval first, and the agent reads the record again after it |
| Agents together | Is there an agent that coordinates other agents? | Yes. They already run in production at LeonAI |
| Out of line | What happens when something breaks the rules? | The agent stops and asks a person. A price off the price list or above the credit limit waits for approval |
| Access and audit | Who sees what, and what remains after the chat? | Access set per employee, one approval per action, and an audit log of every tool call |
| Live or demo | Which agent already works at a customer? | We say which is which. The scenes on our agents page are marked as recorded demos |
| Who builds | Who writes the business rules? | We build the agent, and define the rules and approval points with you |
| Cost | What does it cost beyond the licence? | An agent has no single price list. MCP for Priority has public prices, and the rest of the scope is defined together |
How it works
- One processWe start with one process running end to end
- Rules and approval pointsDefined together: what the agent does and when it stops
- Read onlyThe agent reads and drafts; the team learns to trust it
- Once the team trusts itWrites, one action at a timeEvery action is shown for approval before it is recorded
- More agentsAnd one that coordinates them, like those in production
Does the agent only answer, or also write in Priority?
An agent that only answers saves a search. An agent that also records in Priority, such as a quote or an order, saves the work itself, so it needs brakes. Ask what gets recorded without a person seeing it, and how you know a write really happened.
Our agents work through MCP for Priority. Writing is off by default, and turned on for the users who need it. In a conversation every action is shown for approval first, and after a write the agent reads the record again and reports only when there is evidence. Every check is listed in the guide "How to build an AI agent Priority can trust".
How many agents can work together?
One agent that does everything is hard to check. It is easier when each agent has one clear job: what it writes in Priority, and when it stops and asks a person. With several agents like that, something has to coordinate them.
Agents that coordinate other agents already run in production at LeonAI itself. Companies and organizations already use our Make connector for Priority. On our agents page you can watch a recorded working day, hour by hour: what each agent read, what it wrote and who approved.
What does the agent do when something is out of line?
This is where an agent differs from rule-based automation. Ask the vendor what the agent checks before it prepares the work, and what stops it.
In the field-sales agent we describe in a separate guide, for example, the agent checks the customer's credit limit, the payment terms and the minimum order before it prepares a quote. A price off the price list or above the credit limit stops for a person to approve.
Which agent already works at a customer?
A demo shows what can be built. An agent in production shows what holds up in daily work. Ask the vendor which of the things it shows runs at a customer, and which is a demo.
We write the difference on the page: the scenes on our agents page are marked "Recorded run · demo data", and the agents that coordinate other agents run in production at LeonAI. Companies and organizations already use our Make connector for Priority. We don't publish customers' names without their permission.
And soon you can run an agent yourself. In our Live Lab, as each experiment opens, the agents will work on a real Priority system in LeonAI's test environment, on sample inputs and your own test material, and no record will be written until you approve it. Until then, entry is by invitation.
Who builds the agent, and who sets the rules?
Most of the work in an agent is not the model but the connection to your systems and the business rules: what the agent may do, when it stops and who approves. Ask who writes those rules.
We build agents: systems that do work inside your systems, not just talk about it. The process, the rules and the approval points are defined with you, and if a simple chatbot turns out to fit you better, we'll tell you straight.
When the process itself is not mapped yet, you can start with an audit before building an agent: interviews with the people actually in the role, end-to-end process mapping, opportunities ranked by return on investment, and a prioritized work plan, not a shelf document.
What does an AI agent for Priority cost, and how long does it take?
An AI agent has no single price list, because most of the work is the connection to your systems and the business rules. The technical foundation, MCP for Priority, has public prices on its product page, and the rest of the scope is defined together. Ask any vendor what the licence includes and what is priced separately.
The connection itself is quick: once the API user in Priority is ready, MCP for Priority connects in minutes. What takes time is defining the process, the rules and the approval points. So we start with one process running end to end, and expand from it.
Common questions
- What should I ask to see before choosing an AI agent company?
- Three things in practice: a preview of a write before the approval, a line in the audit log, and an answer that says what time its data is from. And ask which of the things shown runs at a customer, and which is a demo.
- What does an AI agent for Priority cost?
- An AI agent has no single price list, because most of the work is the connection to your systems and the business rules. The technical foundation, MCP for Priority, has public prices on its product page, and the rest of the scope is defined together.
- How long until an agent runs?
- The connection itself is quick: once the API user in Priority is ready, MCP for Priority connects in minutes. What takes time is defining the process, the rules and the approval points. So you start with one process running end to end, and expand from it.
- Is one agent that does everything better than several agents?
- Several agents, each with one clear job: what it writes in Priority and when it stops and asks a person. An agent that coordinates them divides the work. Agents like that already run in production at LeonAI itself.
