How do you keep large test suites from overwhelming AI coding agents?

A developer is seeking strategies to manage a large test suite of approximately 35,000 backend and frontend tests, aiming to reduce maintenance effort while maintaining regression coverage. They are looking for advice on how to prevent these extensive test suites from overwhelming AI coding agents. Specifically, they want to know how to decide which tests to consolidate or remove, keep code searches focused on production logic, and provide AI agents with only the relevant tests without loading the entire suite.