Fixes for edge cases with welcome answers, names, mistaken selections, and Haiku replies showing internal reasoning

Your agents can now use earlier answers to decide when a long conversation is complete or an outreach call needs follow-up. This includes multilingual conversations, such as those in Hebrew, Arabic, Hindi, and Japanese.

Sensei, our AI conversation quality coach, found these problems by reviewing completed conversations and production signals.

• Fixed cases where a text answer sent from the welcome screen disappeared if the chat failed to open.
• Fixed cases where text conversations failed to save a participant's name or a correction, including names given during signed-in conversations.
• Fixed cases where the agent repeated a mistaken selection as fact instead of asking again after the participant withdrew it.
• Fixed cases where Claude Haiku 5.5 text conversations showed participants internal reasoning.