Every engineering leader is under pressure to show measurable efficiency from AI tooling. For web and mobile, that story is becoming familiar. For connected devices — TVs, set-top boxes, consoles, automotive — it is not, because an agent writing code cannot see what happens when that code runs on real hardware.

In this session we close that loop live. We take a real connected-app spec, hand it to an agent on stage, and let it build while we talk through where agentic development for device software is actually working, how teams are adopting it, the failure modes — chief among them never touching a real device until it is too late — and what we have seen work with streaming platforms and device makers.

The agent keeps working in the background throughout. At the end we run what it built across a range of real devices, live, and see how it did.

A working session with a visible result: real code, real devices, in fifteen minutes.