While it builds, the chat lists its steps as it goes: reading your request, designing the flow, building it, testing it on sample data, fixing what the test found, testing again.
Two AI models work on your request at once, and the first good answer is used, so a simple flow comes back sooner.
A small change to a flow is quicker too: the AI writes only the blocks it changes.
A hello on a new, empty flow gets a friendly reply and an idea of what to ask.