Not every build request is the same. Sometimes you know exactly what you want and just need it done. Sometimes you have a rough idea and would rather the AI check its understanding before writing code. And sometimes you're building something complex enough that you want to review the full plan before anything starts. Caffeine now has a mode for each of those.
At the bottom of the chat input, you'll find a new mode selector. Three options.
Instant is the default. Send your message, Caffeine builds. No questions, no back-and-forth. Use it when your request is clear, when you want to move fast, or when you'd rather just see a result and iterate from there.
Thinking asks at least one round of clarifying questions before it builds. If your message has real decisions buried in it — who can access what, how should edge cases work, what happens when something conflicts — Thinking surfaces those before the build starts rather than after. The first result tends to land closer to what you actually wanted.
Pro goes further. Up to three rounds of questions, and Caffeine won't start building until you tell it to. Once it's finished asking, it signals it's ready and waits. You get a chance to review its full understanding of the scope before any code is written. For complex features where a wrong first build costs time, this is the mode to use.
You can switch modes between any two messages, and if you ever want to skip the questions mid-conversation — just tell it to build and it will. The mode is a default, not a constraint. Credits are the same across all three.