With usage-based AI, poor context no longer wastes only time. It can also consume credits while producing weaker answers.

In this rapid-fire session, I’ll show how oversized prompts, entire-repository context, long-running conversations, and unnecessary model changes quietly increase cost and reduce accuracy. You’ll see practical habits for narrowing the working set, defining what “done” looks like, matching the model to the task, and recognizing when a conversation has become too stale to save.

Better results do not always require a more powerful model or a longer prompt. They often begin by removing everything the model does not need.

Takeaways
- Remove irrelevant context before every request.
- Match model cost to task value.
- Reset stale conversations before they drift.