I feel like “streaming” will be more and more of a concept in 2026. We can stream HTML, but rarely do. With AI APIs these days, we’re seeing streaming much more commonly. We want to see that LLM kick out an answer word-by-word, as the perceived performance is better that way. What about a JSON response, can we stream that? Seems harder since JSON needs to be fully valid to be decoded. Krasimir Tsonev shows how it can be done.
