What should I watch out for when setting up BYON in Azure AI Foundry?

Start with the outbound path for agent tool calls. If traffic needs to leave through a hub firewall, plan that egress explicitly rather than assuming the virtual network setup handles it for you.

Also map the Foundry resource dependencies and how each one is reached privately. The agent's network path and its dependencies deserve separate attention.

Don't count on virtual network flow logs to show every private agent dependency call. Foundry's Data Host Proxy affects what those logs can see, so a lack of traffic there doesn't necessarily mean the dependency isn't being used.

There's a governance piece too. Think through how Foundry accounts will be managed alongside the network resources, especially if you're deploying the setup with Bicep. The account model and the private networking dependencies both need to fit your deployment process.

If private Azure API Management is part of the plan, treat it as a separate integration to work through. Authentication, application roles, and rate limiting are all relevant policy questions, in addition to the private connectivity itself.