Multi-Model AI with Philip Kiely

How do you build AI-native software that goes beyond a simple ChatGPT wrapper?

Philip Kiely, software developer relations lead at Baseten, joined me on Software Engineering Radio to discuss multi-agent AI and how to build AI-native software that goes beyond simple ChatGPT wrappers. Philip advocates for composing multiple models and agents that take action to achieve complex user goals—rather than just producing information—and explains what that transition looks like in practice.

We covered the journey from using off-the-shelf models to building custom AI solutions, driven by needs for domain-specific quality, improved latency, and economic sustainability. Philip introduced the engineering challenges of inference engineering—from choosing inference engines to managing GPU infrastructure at scale—while consistently reinforcing that AI engineering is primarily software engineering with exciting new challenges layered on top. The conversation also addressed trust, safety, and observability in agentic systems, concluding with Philip’s recommendation for iterative experimentation as the best path into multi-agent AI.

Multi-Model AI with Philip Kiely on Software Engineering Radio.

Go give it a listen and be sure to share with your colleagues.

Don’t miss all the great episodes of Software Engineering Radio! You can catch the show on Spotify, YouTube (audio only), or by adding the podcast feed to your favorite podcasting app.