Nelson Barringer

This video pretty much summarizes why it feels so hard to switch between LLM agents after getting used to one.

From the video contents:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
THE INSTRUCTION BLOCK — TEMPLATE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
One page, six headings. The first five are stable and describe how you work. The sixth is
current and gets its own section so you can replace it without touching the rest.

1. WHAT I BUILD
   The domain, the stack, the kind of work I actually do day to day.
2. WHAT I ALREADY KNOW
   The heading people skip — and then wonder why they get explained things they teach.
   Name the level, not the modesty.
3. HOW I WANT ANSWERS SHAPED
   Length, format, code-first or explanation-first, how much hedging you want back.
4. STANDING CORRECTIONS
   Anything you’ve had to fix twice. If you corrected it twice, it belongs here.
5. WHAT NOT TO DO
   Negative constraints. Load-bearing, and the ones a new tool has no way to guess.
6. WHAT I’M WORKING ON NOW
   Current only. Replace this section on its own; leave 1–5 alone.

Twenty minutes to write the first time. It is exact text you authored, which is precisely
why it copies into any assistant and still means what it meant.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

<END TRANSMISSON>