I could see the value of using burp as a way to have a GUI look at what your agent is doing, but I know at Vulnetic and other ai security vendors we use our own proxies or custom scripting rather than burp.