Not every connection between AI and your business software is useful. Some save time; others add risk without a benefit. This article explains which integrations can make sense for a business, how to keep permissions limited, and when you are better off not connecting a system.
A useful integration should save a noticeable amount of work and be possible to configure safely. The table shows systems worth considering for an AI integration and the benefits to test for.
| System | What AI does | Benefit |
|---|---|---|
CRM | Look up and update contact information during a conversation | Faster customer follow-up |
CMS | Edit, review, and publish content | Less manual content work |
Project management | Automate status updates and reminders | Help keep tasks from falling through the cracks |
Document sources | Find and summarize information | Get an answer without searching manually |
Do not connect everything. Avoid integrations with more permissions than necessary, systems you rarely use, and connections without logging or oversight. Always ask whether the benefit justifies the added risk.
Give each integration only the permissions it needs. A CRM connection that only reads contact information does not need write access. For a CMS connection, agree separately who may edit and publish, and which changes need advance approval. In our own Storyblok workflow, the agent sometimes publishes directly; we review the result afterward and correct it where needed.
An employee asks Hermes to edit a product description. Hermes reads the current copy from the CMS and prepares an improvement for review. In a draft-first pilot, the employee approves it before publication. Reading, editing, and publishing are separate permissions; restrict each to the agreed workflow rather than assuming that approval itself changes the access rights.
For more on how we set up integrations, see our page on AI automation. To find suitable processes, read our article on automating business processes.
Software with an API or a source Hermes can read may be suitable: a CRM, CMS, project management tool, or document source. We assess usefulness, permissions, limitations, and safety for each situation.
Yes, where it is needed. Agree in advance which edits and publications require approval and which the agent may carry out directly, with human review afterward. Limit access to that scope.
Minimum permissions, logging, and human oversight for high-risk actions reduce risk, but they do not make every integration safe. Assess the data and possible consequences before connecting a system, and review the security and privacy requirements.
