What is an API Pagination?

Two common patterns dominate: offset-based pagination (request page 2 of 50 results per page) and cursor-based pagination (request everything after a specific opaque token marking your place in the list). Cursor-based approaches tend to be more reliable when the underlying data changes between requests, since offsets can skip or duplicate records if items are inserted or deleted mid-scroll. Any client consuming a paginated API has to loop — request a page, check whether a "next page" cursor or offset exists in the response, and keep requesting until it doesn't — rather than assuming a single call returns the full dataset. Forgetting this is a classic integration bug: code that works fine in testing with ten records silently truncates results once a real account has ten thousand. For AI agents this matters directly: an agent asked to "summarize all my emails from this month" needs to actually page through the full result set rather than stopping after the first page and confidently reporting on an incomplete slice of the data.