How to Use ChatGPT Codex for Everyday Work

TL;DR

ChatGPT Codex can complete practical work directly with files, folders, spreadsheets, PDFs, images, websites, and connected apps, even if you do not code. Start with an isolated project folder, choose approval-based permissions, provide a specific prompt, and use an AGENTS.md file to preserve project instructions and working preferences.

Transcript

This is chat GPT codeex. Right now it's one of the most powerful AI tools available and I know it has the word code in it and it did start as a coding app but it could do a whole lot more now and just about anyone watching this video is going to be able to use it. You don't actually need to know anything about coding in order to get a lot out of th... Read More

Key Insights

• Codex is an AI agent that can perform knowledge work and coding tasks within the same conversation. Unlike a chatbot that only supplies an answer, it can inspect files, create deliverables, and carry out requested work on the user's computer.
• Projects are folders that isolate related inputs, instructions, and outputs on the computer. Keeping each job in its own project reduces cross-contamination, makes created files easier to find, and limits the initial scope of the agent's file access.
• Permission settings determine which files, internet resources, and actions Codex can access. The recommended beginner setting allows it to work within the project while requesting approval for external files, edits, internet use, or actions considered potentially unsafe.
• Clear prompts improve efficiency by identifying the source location, required fields, desired output format, and destination. If Codex can access a wider portion of the computer, explicitly naming the folders it should inspect can help it work faster and reduce usage.
• Receipt processing works by reading images, extracting structured details, and generating a spreadsheet deliverable. In the demonstration, Codex collected vendor names, dates, totals, and categories, then created an Excel workbook with a dashboard showing spending by category.
• AGENTS.md is an onboarding document that gives Codex reusable context and behavioral instructions. It can record details such as the user's occupation, spreadsheet color-coding preferences, preferred programming language, working style, and information specific to one project.
• Codex usage is credit-based even though the app is described as available to everyone for free. Higher reasoning effort and faster processing can consume more credits, while standard speed with high reasoning effort is presented as a lower-usage configuration.
• Codex can extend beyond local project files through plugins, computer use, and connections to apps. The described examples include Gmail, Slack, Google Drive, Chrome, Canva, websites, PDFs, images, and spreadsheets, although broader access should be granted carefully.
• More videos from this channel:
  • How to Use ChatGPT 6 Astra for Better Work
  • How to make cinematic YouTube videos with LTX Studio
  • How to Use ChatGPT for Real Work Without Coding
• More from this channel: Skill Leap AI

Questions & Answers

Q: How can beginners use ChatGPT Codex without coding?

Beginners can create a project folder, place relevant files inside it, and describe the desired result in ordinary language. Codex can then read those files and create outputs such as spreadsheets, dashboards, cleaned folders, or websites. The receipt example required no programming: the prompt simply identified the fields to extract and requested an Excel dashboard organized by spending category.

Q: How should a new Codex project be organized?

A new Codex project should normally be created from scratch as a dedicated folder on the computer. Inputs for one assignment can be copied into that folder, and Codex can save its outputs there as well. Keeping projects isolated prevents unrelated work from becoming mixed together, makes deliverables easier to locate, and lets users begin with a limited file-access scope.

Q: What permission setting should Codex beginners choose?

Beginners should use the approval-based option that lets Codex work in the selected project while asking before potentially unsafe actions. The transcript advises against unrestricted access for someone following a beginner guide. Full access can let the agent reach the internet and files across the computer, but the narrower setting provides more supervision while users learn how agent actions work.

Q: How does Codex turn receipt images into an Excel dashboard?

Codex reads the receipt images stored in the project folder and extracts the requested information, including vendor names, dates, totals, and spending categories. It then places the structured data in an Excel workbook and builds a simple dashboard that summarizes spending by category. In the demonstration, the resulting workbook appeared in an output folder, alongside a PNG image of the dashboard.

Q: How can prompts make Codex faster and use fewer credits?

Prompts should clearly state where Codex should look, which information it should collect, what processing it should perform, and the exact deliverable to create. This becomes especially important when the agent can access many folders across the computer. Naming the relevant location and output requirements reduces unnecessary searching, and the transcript states that greater prompt clarity can bring usage down.

Q: What is an AGENTS.md file in ChatGPT Codex?

An AGENTS.md file acts like an onboarding document for the agent. It stores instructions and context that Codex can read when starting work, so users do not need to repeat the same preferences in every conversation. It can describe the user's role, project background, spreadsheet formatting rules, preferred coding language, and working style, either for one project or more generally.

Q: How do model, reasoning, and speed settings affect Codex?

The selected ChatGPT model drives the agent, while reasoning effort controls how much work it applies to complex tasks. The transcript recommends high reasoning for typical use and notes that extra-high reasoning consumes more credits. Speed also affects usage: fast processing is described as costing more credits, while standard speed with high reasoning is suggested for keeping consumption lower.

Q: What everyday tasks can ChatGPT Codex perform?

Codex can work with files, folders, spreadsheets, PDFs, receipts, images, websites, and connected apps. The examples described include converting receipts or PDF invoices into organized spreadsheets, cleaning a messy folder, renaming and sorting files, creating websites, and working with Gmail, Google Drive, Chrome, Canva, or Slack. Plugins, skills, automations, memory, and computer use broaden these workflows.

Summary & Key Takeaways

• Codex is an AI agent that combines conversation with direct task execution. It can inspect source files, create new documents, perform knowledge work, and handle coding within the same chat. Projects keep its work organized inside designated computer folders, while optional connections extend its reach to apps and other locations.

• The receipt demonstration shows a practical workflow for non-coders. After receipt images were placed in a project folder, Codex extracted vendors, dates, totals, and categories. It then produced an Excel workbook with a spending dashboard, saved the workbook in an output folder, and also created a dashboard image.

• Safe setup depends on matching permissions to the task. Beginners are advised to restrict Codex to the project folder and require approval for potentially unsafe actions. Clear prompts can reduce unnecessary work and usage. Persistent instructions stored in AGENTS.md can define personal context, formatting preferences, coding choices, and project-specific working rules.