LLM Inference Interview Questions #3 - The Context Condensation Trap

Why your agent’s carefully structured plan quietly degrades into mush after 100 steps, and the architecture trick to decouple your source of truth from a medium designed to forget.

You’re in a Senior AI Engineer interview at Anthropic and the interviewer asks:

“Your coding agent uses a task-tracker tool, and the current plan is already in its output. So why do you also write that plan to a tasks.md file on disk, what does that actually buy you?”

Don’t say: “It’s a backup” or “so the user can read it.”

Technically true. Strategically …