A team building an AI agent finds that its answers frequently become inaccurate as source systems update throughout the day. According to the streaming-focused session, what is the core problem being solved by serving continuously refined, low-latency context?