How an agent runs
Every agent we deliver follows this shape. You decide where the approval points sit.
- 1
Trigger
A new email, a form, a schedule or a request from a person starts the run.
- 2
Gather
The agent reads the inputs it needs from the systems you have connected.
- 3
Work
It carries out the defined steps, using each tool only within the permissions you set.
- 4
ApproveYour call
Before any action you flag, the agent pauses and a person confirms or corrects it.
- 5
Report
It delivers the result with a log of what it read, decided and did.
Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. We keep each agent narrow, prove it on your examples, and build the controls in from the start.
Source: Gartner, June 2025
What you receive
- The configured agent and its source
- Documented permissions and approval points
- A runbook for the person who owns it
- The test set used to prove it
- A walkthrough with your team
Questions
Can the agent act without a person?
Only where you allow it. You choose which actions need approval, and you can change that later.
What systems can an agent connect to?
Most systems with an API or a supported connector, such as email, calendars, CRMs, spreadsheets and document stores. We confirm the specific connections on the scoping call.
What happens when our process changes?
The configuration and source are yours, and the runbook explains how to change them. We can also make the change for you.
Which AI models do you use?
The ones your organization has approved. The documentation notes what would need to change if you switch.