We use one model across all three surfaces called Task. Those do have different information architectures, i.e. Linear tickets have correspondence or comments, LLM conversations have chat history and code reviews have diffs and comments. But, at the end of the day, all those information are used to output the correct code and we are doing just that.