Use the Codex app for ongoing non-coding research
The poster uses the Codex app as a stateful way to build knowledge and data over time, processing their department and applying that context to cross-discipline tools.
Done withCodex
- What they did
- The poster says they use the Codex app as a stateful research workspace rather than just for coding, because it can accumulate context and knowledge over time across chats. They also rely on an AGENTS.md file for persistent style and instruction control, and describe switching between coding or desktop operations and detailed question-answering inside the same app.
- How it went
- The reported result is qualitative: they preferred Codex’s answers to ChatGPT’s Research mode and found it useful for lightweight web searches and ongoing research. The main downside mentioned in the thread is that the app still felt bare-bones.
- Worth knowing
- One practical catch from the comments is that Codex usage shares the same overall chat-model quotas across ChatGPT, web, Codex app, and Codex CLI, except for Codex Spark.
Try it yourself with Codex
I want to use you as an ongoing research partner for [my department or topic]. Read the documents in [folder path], build a structured knowledge base in [format] covering key people, processes, data, and open questions, and write down how you organized it so you can pick up where you left off next session. You're done when I can ask you a question about [topic] and you answer from the saved notes with sources, and you should tell me before changing or deleting any of my original files.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · Undated
Related use cases
Turn tickets into merged pull requests
Pulls tasks from GitHub Issues or Linear, then runs Claude Code or Codex in isolated Kubernetes pods to carry each one through to a PR.
Diagnose a killed process on a server
Codex CLI over SSH found the OOM killer in kernel logs on a 2 GB droplet with no swap and listed fixes by risk without changing anything.
Find bugs with property-testing agent swarms
Agent swarms doing property testing found about 30 bugs in the author's work code, then ~100 bugs across open-source projects like uv, with many fixes merged.
Find connections across 100 non-fiction books
Claude Code browses a personal library with CLI tools and surfaces trails of related excerpts across books.
Turn any URL into a structured Notion note
An n8n workflow fetches a submitted URL, strips it to readable content, has OpenAI summarise it into structured fields, and saves a Notion note.
Measure how long jobs stay open at tech companies
An agent read 25 open jobs each at five companies and measured how long each had been posted, e.g. 21 of 25 at Figma were 60+ days old.
Submitted by OpenKrill
Custom agent
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.