Nobody writes a chat message like an email. They send “add to my todo list”, then “gym tomorrow”, then “at 7”. Three messages, one thought. An agent that answers the first piece replies to half a request, then replies again when the rest arrives.
A human would just wait. The hard part is knowing how long, and Telegram gives bots no typing signal, so the bot has to guess. What worked: reply six seconds after the last piece, give an unfinished line up to eight more seconds, and never treat a line that ends on a connecting word as done.
The best part is that it costs nothing extra. The router already reads every message, so asking it whether the turn is finished does not need another model call.
Small waits are product decisions. Users never notice a good one and always notice a bad one.