How to use OpenAI Codex as a non-technical business user

Executive overview

In ChatGPT, you bring data to the AI — files, context, prompts — all held in the model's head at once. As the context grows, performance degrades. Codex flips this: the AI comes to your data, selects only what it needs for each task, and maintains its focus throughout.

This makes Codex 2–3x more effective for complex, multi-step work. Every feature you already use in ChatGPT has a direct equivalent in Codex.

The core shift: AI-to-data beats data-to-AI for anything complex.

Three questions to answer when you first open Codex

• Where should the work live — a general chat or a specific folder on your machine?
• How hard should the AI think — low reasoning for quick tasks, extra high for complex ones?
• How much freedom should it have — start with default permissions, move to full access as trust builds.

ChatGPT vs Codex: feature translation

• Chats — identical in both; same back-and-forth thread interface.
• Projects — in Codex, a project is a folder on your computer; conversations inside that folder share context.
• System prompts — add an agents.md file to a folder; Codex reads it before every interaction in that project.
• Apps / Plugins — Codex plugins bundle an app connection plus a skill (step-by-step instructions for using that app); this makes tool use far more reliable than in the browser.
• Scheduled tasks / Automations — Codex can read and write to connected systems, so automations produce real output (drafted emails, updated dashboards) rather than read-only reports.
• Browser use — a live browser runs inside Codex; the AI can navigate software you don't know how to use and complete tasks autonomously.
• Memory — instead of a limited in-model memory, Codex writes preference files to your folders and searches them when relevant; effectively unlimited.

Managing usage and cost

• Model choice: use 5.5 for complex tasks, 5.4 for lighter work.
• Reasoning level trades speed and usage quota for quality; don't avoid extra high for genuinely complex tasks.
• Monitor usage under Settings → Usage Limits or via the rate limits indicator at the bottom of any chat.
• The $200/month plan rarely hits limits even with heavy use; the $20 and $100 plans require more attention.

Five high-value use cases

1. Updating dashboards and files incrementally — create an Excel sheet or dashboard once; on future weeks, drop new data into the folder and instruct Codex to update only that. Eliminates full rewrites and reduces error risk. Can be automated on a schedule.

2. Bulk file organisation and insight extraction — point Codex at a folder of client or project files; it renames files, removes duplicates, merges data, flags edge cases, and writes a summary of what changed and why — including terminology preferences for future tasks.

3. Browser automation for software you rarely use — log into any tool inside Codex's built-in browser and describe what you need. Example: adding 150 contacts to an email sequence that would have taken six hours manually was completed in 30 minutes without intervention.

4. Answering hard questions across many systems — Codex can pull from local notes, search the web, check your inbox or calendar, and produce a one-page decision summary with sources, assumptions flagged, and open questions for you to resolve.

5. Email integration with proactive briefings — connect Codex to Gmail or Outlook via plugin; set an automation to draft a daily or weekly briefing (what happened, what's coming) directly into your inbox as a draft, readable on mobile without opening Codex.

When to use ChatGPT vs Codex

• ChatGPT in the browser: quick, ad hoc, smaller tasks.
• Codex: complex, multi-step tasks involving files, tools, or multiple data sources where higher reliability matters.