An agent and a CRM can both be running without being able to work together. There is another part to consider: the connection between them.
That distinction is useful when you are choosing tools, setting them up, or trying to understand why a task has stopped. Each part has its own purpose and its own requirements.
The agent carries out the task
An agent receives a request and uses the capabilities available to it. Those capabilities depend on the particular software, model, tools, and configuration.
Two agents may present similar chat interfaces while supporting different actions. Look at the work you need done and the interfaces the agent actually supports. A product name on its own does not describe a complete working setup.
The application holds the work
A CRM organizes customer information. A shared inbox organizes conversations. A knowledge base organizes documents. People can continue using those applications directly.
When an agent uses an application, the application still has its own records, access rules, and state. A draft in the agent’s conversation is not automatically a saved CRM note. A prepared reply is not automatically a sent message.
The connection defines the access
A connection gives an agent a supported way to use an application. It needs a compatible interface, valid authorization, and a clear scope.
| Intended task | Information needed | Access to consider |
|---|---|---|
| Summarize a company | Selected CRM records | Read the relevant records |
| Save a meeting note | The destination record | Write to that record |
| Prepare a reply | A conversation and relevant policy | Read the selected context |
| Send a reply | The intended conversation | Separate permission to send |
These are examples for thinking about access. The actual options depend on the products and the connection they support.
Readiness belongs to each part
If the application is available but authorization has expired, the connection may need attention. If the connection works but the selected model is unavailable, the agent’s task may still be unable to proceed.
Good status information helps you locate the problem. It should tell you whether the issue concerns the application, the connection, or the task, and what action would help.
An app being available and an agent being able to use it are two things to check.
Begin with a clear purpose
Choose the task first. Identify the context it needs. Then look for a combination of agent, application, and supported connection that can carry out the work.
This is the relationship Peren is designed to make easier to understand. The examples on this site illustrate that direction; they are not evidence of live integrations.