The first wave of enterprise AI agents was measured by the wrong metric: how many tickets they deflected. The next wave will be judged on whether they actually resolve the customer's problem and whether they improve on their own once they're live. That shift, from launching an agent to running one that improves itself, is where most pilots stall and where the real frontier sits. In this fireside, Decagon President and Co-Founder Ashwin Sreenivas joins CNBC's Arjun Kharpal to map where agentic AI in the enterprise is actually heading, by drawing on millions of production conversations across companies like Chime, Oura, and Hertz, and on the hard parts no demo shows: trust, governance, and keeping humans in control as agents start to improve themselves.