in my experience claude’s “research” mode isn’t particularly helpful. i found its reports consistently less useful than normal conversations. haven’t tried comparing to claude code with intermediate artifacts yet. does this track? has anyone had much better experience with research mode?