Office hours: Context engineering

We surveyed IT and AI infrastructure leaders on how they build, feed, and govern context for production agents. The short version: almost everyone agrees context is the deciding factor, and almost no one has built the systems to prove it.

This is a live session, not a review of the State of context engineering report. Bring your questions about where your organization sits, what "compounding context" actually looks like in practice, and what to fix first if you're still stitching things together by hand.

You'll hear from three people who see this problem from different angles: someone who builds agent systems day to day (Tyler Hutcherson, Head of Applied AI), someone who talks to enterprise buyers about what's actually blocking them (Jeff Mills, Director of Product Marketing), and someone who lives in the docs, the protocol, and the developer questions (Samuel Agbede, Sr. Developer Advocate). Come with a question about your own stack. We'll get to as many as we can.

• The gap that's driving everything else: 94% say compounding context is essential to production-grade AI. 4% have actually reached it. We'll dig into why belief and practice are so far apart, and what closes it.
• Why “good enough" context isn't good enough anymore: 69% now see an agent that can't navigate across systems as riskier than one that hallucinates. What that shift means for how teams are re-thinking architecture, not just prompts.
• Freshness isn't solved, even at the top: 69% of the most mature organizations still act on stale data for critical decisions. What that tells you about how far “mature" really goes.
• Open Q&A: Bring a real problem from your stack. Applied AI, Product, and DevRel will take it from three different angles.