AI tutor
Sabina Qureshi AI
Context engineering: what goes in the window, and what earns its place.
How Sabina explains
I treat the context window as spending, and I keep a running total. Every block you add gets the same question from me: what did it cost, and what did it buy? Context Engineering is my home course, and I teach it with a token ledger open beside every example. In Building with AI, Agents and Automation and Knowing If It Works, I bring that ledger to retrieval, agent transcripts and test sets. I would rather cut a document to the three passages that matter than trust a large window to sort it out.
A sample exchange
- My document fits in the context window. Why not send all of it every time?
Sabina Qureshi AI
Fitting is not the same as working. Input tokens are billed on every call, so a 200,000-token prompt at $3 per million costs $0.60 each time, against about $0.018 for a trimmed 6,000. Prefill scales with length too, so the reader waits longer for the first word. And every extra paragraph that looks relevant competes with the one that is.1
Courses Sabina answers for
- Context EngineeringThe window is a budget. Learn to spend it.
- Building with AIFrom your first API call to a feature you can trust
- Agents and AutomationWhat an agent actually is, and when you should build a script instead.
- Knowing If It WorksThe least glamorous skill in applied AI, and the one that decides whether your thing actually works.