How we keep streams alive just long enough

How we keep the runtime alive when processing streams at 14.ai: using Effect Stream + Deferred.

Every customer conversation is not just streamed to the user, it is also tracked in parallel so we can power downstream workflows like summarization, triaging, urgency alerts, and insights. To make sure all the information is collected to kick off this task, even after the customer's response has been sent, we keep the system alive just long enough to finish processing everything behind the scenes.