Model Ecosystems and Harness Design for Cost-Effective Scaling of Enterprise Agents

As enterprises move from experimenting with AI to deploying agents that do real work, a quiet shift is reshaping where performance comes from. The model is no longer the whole story. The harness around it — the system that decides what context the model sees, which LLMs take on which tasks, and how work gets structured — is now a dominant lever for performance.

This masterclass gives attendees a structured framework for designing harnesses and investing in model ecosystems, drawing on Glean's journey building an agentic platform for the enterprise. We'll see how a diverse model portfolio and a well-designed harness play crucial roles in a platform that can support the full spectrum of realistic enterprise workloads at sustainable costs.

The session also examines when it makes sense to invest in specialist models for high-leverage tasks, illustrated by Waldo, Glean's agentic search model — built in collaboration with NVIDIA using Nemotron 3 Nano and optimized for fast, cost-effective enterprise context retrieval at scale.

What you'll learn

• The Context–Harness–Model framework: How to prioritize investment across the three highest-impact determinants of enterprise agent performance
• Harness design for token efficiency: How to design harnesses that optimize for both efficiency and accuracy
• Building a model ecosystem: How multiple models work together to deliver frontier performance at manageable costs
• Platform performance in practice: How these principles come to life in real-world enterprise use cases

Who should attend:

AI platform leaders, technology strategists, and product leaders