Only 1 in 1,600 People Use Codex. Here's How to Catch Up.

TL;DR

• 1

  Codex changes the unit of work: Nate says his token usage exploded to 300 million to 500 million tokens a day because he stopped asking for answers and started assigning whole jobs like reading folders, rendering documents, using websites, and checking outputs.

• 2

  The real pattern is a 'chief of staff thread': Instead of scattering work across random chats, he keeps one persistent thread pointed at a goal, its sources, artifacts, standards, and next steps so the AI becomes a home base for the project.

• 3

  Goals beat prompts: Codex gets much better when you define the objective, sources, quality bar, permissions, and proof of completion, then tell it not to stop at the first plausible draft.

• 4

  Skills make corrections compound: A one-off fix is just a chat, but turning repeated corrections into reusable skills, checklists, or workflows creates a system that improves over time across documents, research, dashboards, and support work.

• 5

  A personal work dashboard is a killer example: He describes building a custom heads-up display that watches sources like Slack, email, and browser-based tools, reprioritizes what matters every 15 or 30 minutes, and surfaces the work that needs attention.

• 6

  Power needs boundaries: Nate stresses using .env files for secrets, limiting write or publish permissions, and requiring receipts like logs, renders, tests, and command output so agentic work stays inspectable instead of turning into hype.