Free Lesson
Context Engineering: Budget the Window Like Memory
30 min
Aug 18, 2026 1:30 PM
Virtual (Zoom)
In this video
What you'll learn
Measure what your tool definitions really cost
Count the tokens your MCP servers and schemas consume on every request before any work happens.
Design an eviction policy for a long trajectory
Compaction, summarisation and retrieval, chosen against what each one loses and when it runs.
Protect the cache prefix that pays for itself
What invalidates a cached prefix, what that costs per turn, and how to order context so it holds.
Why this topic matters
Context engineering is the term everyone now uses and few people do arithmetic on. Several MCP servers can eat a tenth of the window before a single useful token is spent, and that cost repeats on every request you serve. Treating context as an allocated resource with an eviction policy is what separates a demo from a system you can afford to run. You will leave able to price your own setup.
