AI development is generating a lot of buzz, but the reality for teams actually deploying these systems is quite different. Conversations with CTOs and product leaders revealed that while many companies are still working on basic Gen-1 AI experiences—simple input and output systems—only a handful are innovating with advanced Gen-2 AI. These teams are creating more interactive, long-running agents that need to manage user interactions over time, which is where most of the volume lies. They face significant challenges, particularly in user experience (UX), rather than just model performance. A key issue is the lack of standards for agent-to-client communication. Many teams start with basic HTTP protocols but end up creating complex, custom solutions. This can lead to fragile systems that break under real user conditions. Users expect clear visibility into what the AI is doing. If they don't see progress, trust erodes. Teams are investing in ways to signal activity and manage interruptions effectively since users want to steer tasks and correct mistakes in real-time. Continuity is another vital aspect; users expect their AI to remember past interactions, regardless of the device or session. This demand for seamless experience adds complexity, particularly in customer support scenarios where multiple tasks may be happening simultaneously. Teams are also grappling with frequent failures at scale—partial responses and interruptions are commonplace. Cost considerations are starting to shape AI operations, with teams rethinking when and how to run costly processes based on user presence. As these challenges mount, companies are realizing they need to focus not just on AI intelligence but also on creating a robust infrastructure that supports user interactions effectively.