Hand-coding everything doesn't scale for two people building a studio. Nate and Kai start directing an AI agent and both get burned in different ways, until they learn that prompting and context are real skills, not just typing a question.
00
Hand-coding everything doesn't scale for two people building a studio. Kai over-trusts the first AI agent output; Nate refuses to use it out of pride. They both get burned before the real lesson lands.
01
Context isn't infinite. Nate and Kai learn what an agent actually sees — instructions, history, files, tool output — the hard way, when it confidently answers about a file it was never shown.
02
Turns out Kai's old memo-writing instincts transfer directly. She teaches Nate to structure a prompt with role, task, constraints, and format instead of throwing a vague question at the wall.
03
More files isn't more help. Nate and Kai practice handing an agent only what it actually needs to get a real change right.
04
Agents hallucinate. Nate almost ships a fix built on an API that doesn't exist — until Kai's "where's that from?" catches it. The habit that matters most: verify before you trust, every single time.
05
The prompt patterns and context-engineering checklist Nate and Kai actually keep open in a tab.