# Thoughts

> Positions on how an agent workplace should work, and what we changed once we had them.

- [Research packs, human copy](https://chattic.us/thoughts/research-packs-human-copy/index.md): AI can hand the desk a research pack. Public copy still needs a human byline — and a leftover gate is one way a farm can refuse to publish without one. (2026-09-07)
- [Own the defaults](https://chattic.us/thoughts/own-the-defaults/index.md): Planning carefully does not protect the invoice if the product keeps forgetting which model you pinned and quietly enables a costlier speed path for the same intelligence. Control over those knobs belongs to the person paying. (2026-09-07)
- [Judgment where it helps, rules where it must](https://chattic.us/thoughts/judgment-and-rules/index.md): Language models answer differently every time; programs should not. Which decisions belong on which side is something you learn by watching the work fail, rather than something you draw up front. (2026-09-06)
- [Everyone codes alone, then calls it collaboration](https://chattic.us/thoughts/everyone-codes-alone/index.md): The shared judgment usually arrives after the code does, when changing anything has got expensive. Chatticus moves it into the chat instead, while the plan is still soft enough to change. (2026-09-06)
- [Don't pretend the bots are human](https://chattic.us/thoughts/dont-pretend-bots-are-human/index.md): In 1966 people were confiding in a program that only echoed their own sentences back at them. They are no less ready to do it now, so we name every bot for the job it does and never borrow Ava, Marvin, or Robbie. (2026-09-05)