The end of token-maxxing? Scale AI with less spend | BRKSP612
Vignesh Sridhar presents model routing as a way for engineering teams to balance growing enterprise AI usage with rising costs. The Microsoft Ignite 2026 session focuses on software-development workloads and includes a customer example and a live Fireworks Nexus demonstration.
Task- and cache-aware model routing
The session explores how intelligent routing directs coding requests to the right open and closed models, with the goal of reducing token spend while maintaining code quality.
The coverage includes:
- Task-aware routing: Selecting models according to the software-development task.
- Cache-aware routing: Incorporating cache awareness into routing decisions.
- Open and closed models: Routing requests across both model types rather than treating every request the same way.
- Real coding workloads: Using a customer example and a live Fireworks Nexus demo to illustrate the approach.
The central concern is the balance between cost and quality: controlling token spending without undermining the usefulness of AI-assisted development.
Session details and resources
- Speaker: Vignesh Sridhar
- Event: Microsoft Ignite 2026
- Session: BRKSP612 | Data
- Format and level: Breakout | (200) Intermediate
- Session learning resources