Everyone is talking about AI agents, but most of the attention is still on the model.

We think they're looking in the wrong place.

The real engineering challenge isn't getting an AI to think. It's deciding what it can remember, what it can access, what it can do, and how you keep a human in control.

That's the harness, and it's where trustworthy agentic applications are built.

As AI systems evolve from simple assistants into autonomous, multi-agent applications, the complexity shifts from prompting to engineering. Memory, permissions, tool calling, workflows, approvals, provenance and observability become just as important as the intelligence of the model itself.

In this session, we'll explore the architectural patterns and practical engineering decisions behind building trustworthy agentic systems. We'll examine how specialist agents can collaborate, share context, delegate responsibilities and operate within clearly defined boundaries, without sacrificing security, transparency or human oversight.

Through practical scenarios, architectural examples and lessons from emerging agentic development patterns, we'll unpack what makes an effective harness, where things can go wrong, and how to design systems that remain understandable and governable as their autonomy increases.

Because the future of AI engineering isn't just about better models or cleverer prompts. It's about better harnesses.

You'll leave with practical patterns, design principles and a clearer understanding of how to build agentic applications that are useful, transparent, governable and worthy of your users' trust.