Context engineering for analytics agents that actually work

Setting up Agentic analytics is a must for every data team in 2026. But the blocker is often: which tool should I use? How should I configure my agent so it's actually reliable?

I ran many studies on context engineering, evaluating what improves most the reliability and performance of analytics agent - metadata, data profiling, business rules, dbt docs, MetricFlow semantic layer, etc. Each experiment evaluated the agent on 4 KPIs: reliability, coverage, cost, and speed.

In this session, I'll walk you through:

• Why data teams need to become context engineers
• Which context layers actually drive agent performance (and which don't)
• How to change data modeling rules to fit agents needs
• How to build a MetricFlow semantic layer that works for agents

You'll leave with a reproducible methodology to set up a reliable analytics agent on your data stack stack - not a demo, but a framework you can apply to your own data.