Codex 101 for Academic Researchers (Bilal)

  • Author/Source: Mushtaq Bilal, X thread

  • Original: https://x.com/MushtaqBilalPhD/status/2054565093430493609

  • Key Ideas

  • The Codex counterpart of Claude Code 101. Same framing, same target audience: academic researchers with no technical background.
  • Codex is described as a desktop app that runs on your computer, opens a project folder, and works "inside" your files — the same "bring AI into the folder" mental model as Claude Code.
  • Codex equivalent of CLAUDE.md is AGENTS.md: a visible instruction file Codex reads when starting a session.
  • Use cases identical to the Claude Code 101 framing: qualitative researchers extract themes from transcripts; quantitative researchers clean CSVs, run descriptives, decode reviewer comments.

  • Same caveat: Codex is not a replacement for expert judgment. Same closing point about what counts as argument/evidence being the researcher's responsibility.

  • Summary

A near-parallel of the Claude Code 101 tutorial, retargeted at the Codex ecosystem. The substantive differences from the Claude Code version are small but worth noting: Codex's instruction file is AGENTS.md (not CLAUDE.md); Bilal acknowledges that Codex may need extra checking on scanned/badly-formatted PDFs; and the session-continuity language is slightly more cautious ("can continue with saved sessions and project instructions") than the Claude Code framing ("remembers you").

The value of having both tutorials in the same series is that researchers comparing tools (or organizations that have settled on the OpenAI stack) get the same onramp without retranslating concepts. The framing remains deliberately non-technical and the worked examples are identical, which makes the tools feel more interchangeable than they may actually be in practice.

  • Relevance to Economics Research

Useful as the Codex-ecosystem onramp for economics colleagues who have ChatGPT Plus/Pro subscriptions and prefer not to add a Claude subscription. The series also implicitly endorses Panjwani's recommendation that academics on tight budgets should default to ChatGPT/Codex over Claude — a contested claim worth flagging when sharing with faculty.