ChatGPT Codex for Development Teams

Course overview

Many development teams already use AI to help write code, but mostly for line-by-line completion or questions in a chat window. Codex works as a coding agent: it reads the whole repository, runs commands, edits several files and hands work back as a diff or a pull request. The real question is no longer whether AI can write code, but how a team delegates work to an agent so the results meet its standards, can be checked and do not open security risks.

This course takes development teams through using Codex on a real repository, from the first task to a pull request. Learners install the CLI and IDE extension, set up config.toml, and write AGENTS.md and skills so the agent follows the team's standards. They practise delegating, monitoring and steering tasks mid-run, then connect the work to Git and GitHub through branches, code review and parallel tasks in the cloud, while setting the sandbox, permissions and network access to suit the organisation. The course compares Codex with Claude Code to help choose the right tool and closes with a workshop on the team's own backlog. (2 days, 6 hours per day, 12 hours in total, Intermediate to Advanced level.)

What you’ll gain

Explain how a coding agent differs from code completion and choose tasks that suit an agent
Install and configure the Codex CLI, IDE extension and config.toml on a real repository
Write AGENTS.md and build skills that keep Codex working to the team's standards
Delegate, monitor and steer an agent's work mid-run in a systematic way
Take work from a branch to a pull request with tests, and use Codex to review code on GitHub
Run several tasks in parallel in Codex cloud and accept only the results that pass
Set the sandbox, approval mode and network access safely in line with company policy
Compare Codex with Claude Code and plan how to bring a coding agent into the team

Who this course is for

Software and full-stack developers who want to use Codex on their team's real work
Tech leads and engineering managers who need to set standards for using a coding agent across the team
DevOps and platform engineers who manage sandbox permissions and the Codex connection to GitHub
Teams already on ChatGPT Business or Enterprise who want to get full value from Codex
Developers who use AI for code completion and want to move on to working with an agent

Prerequisites

Fluent in at least one programming language; this course does not teach programming
Comfortable with Git, including branches, commits, merges and pull requests on GitHub
Able to use the command line and VS Code or another editor that supports Codex
A personal or company ChatGPT account on a plan that includes Codex, and your own GitHub account
A laptop where you can install software, and a repository you are allowed to use with AI, or the sample repository

Curriculum

Course Details

This course focuses on Codex and how a development team adopts it. It runs for 2 days, 6 hours per day (12 hours in total, 09:00-16:00), as lectures with labs on a team repository that learners are allowed to use, or on a sample repository. Intermediate to Advanced level. Learners must already write code and use Git; programming is not taught. Learners use a personal or company ChatGPT account on a plan that includes Codex and their

own GitHub account; the institute does not provide accounts. It differs from AI Coding Agent for SDLC, which covers coding agents across the whole software lifecycle without tying to one tool, and from Claude Code Roadmap, which covers Claude Code in depth. Here the focus is Codex itself: AGENTS.md, skills, config.toml, pull requests, GitHub code review, Codex cloud, sandbox, permissions and team security policy. Codex menus and features change often, and the content follows the latest version on the training day.

Day 1 Setting Up Codex to Work to Team Standards

Section 1: How a Coding Agent Differs from Code Completion

Code completion fills in code piece by piece, while an agent reads the whole repository, runs commands and edits many files
The agent loop: read context, plan, act, run tests and report back
An overview of where Codex runs: CLI, IDE extension, the ChatGPT app, Codex cloud and GitHub
Tasks worth delegating to an agent, tasks the team should keep, and limits to be aware of

Section 2: Lab: Installing and Configuring Codex on a Real Repository

ChatGPT plans that include Codex compared with using an API key, and what each one covers
Install the Codex CLI and IDE extension and sign in with a ChatGPT account
Ask Codex to explain the project structure, how to build it and how to run the tests
Lab: have Codex fix one small bug, then read the diff before accepting the change

Section 3: config.toml, Models and Sessions

The layout of ~/.codex/config.toml, project-level settings and choosing a model
Common slash commands, and connecting external tools through MCP
Pick up earlier work with codex resume, and start a new session when the context grows too long
Lab: set up separate profiles for editing code and for read-only code exploration

Section 4: AGENTS.md: Standards the Agent Reads Every Time

How Codex combines AGENTS.md from the user level, the repository root and subfolders in a monorepo
What to write: build and test commands, code style, folder layout and things that must not be done
Keep rules short, clear and checkable, and never put secrets in the file
Lab: write AGENTS.md for the repository and test that Codex actually follows it

Section 5: Skills for the Team's Repeated Work

A skill is a folder with a SKILL.md file plus scripts and reference material
Repository and user-level skills, sharing them through Git, and invoking them by name or letting Codex choose
Deciding what belongs in AGENTS.md and what should be a skill
Lab: build a skill for routine work, such as adding an API endpoint the team's way

Section 6: Lab: Delegating, Monitoring and Steering Tasks

Write a task brief: goal, scope, relevant files and a definition of done
Have Codex plan before acting, and review the plan before allowing code changes
Monitor the work, step in when it drifts, and split large work into checkable pieces
Lab: delegate a medium-sized feature and steer it at least once along the way

Day 2 From Pull Request to Organisation-Wide Use

Section 7: Lab: From Branch to Pull Request

Have Codex work on a separate branch and write commit messages in the team's format
Have the agent write and run tests before reporting the task as done
Check the work with /review before committing, and read every diff yourself
Lab: turn the Day 1 work into a pull request with a clear description on your own GitHub

Section 8: Code Review with Codex on GitHub

Connect GitHub to Codex and enable automatic reviews, or request one in a PR with @codex review
Write the team's review rules in AGENTS.md so Codex checks against them
A person still approves: set up the review and merge flow together with CI
Lab: open a PR with deliberate mistakes and compare the Codex review with a human review

Section 9: Lab: Running Parallel Tasks in Codex Cloud

Set up a cloud environment: repository, setup script and required variables
Run several tasks at once in the cloud, or use Git worktrees locally without conflicts
Check the diff and test results, then open a pull request or bring the changes back locally
Lab: send three backlog tasks in parallel and accept only the results that pass

Section 10: Sandbox, Permissions and Network Access

How the read-only, workspace-write and danger-full-access sandbox modes differ
Choose a mode with /permissions, admin-enforced requirements, and when not to use Full access
Network in the cloud: off by default, domain allowlists and limited HTTP methods
Lab: try each mode against risky commands, then write a standard config.toml for the team

Section 11: Security, Policy and Choosing the Right Tool

Prevent secret leaks: .env files, tokens and customer data the agent should not see
Never run an agent with full access on production machines, and review every diff before merging
Licensing and intellectual property of AI-generated code, and the company's AI policy
Codex compared with Claude Code: context files, permission settings and where each one fits

Section 12: Workshop: Capstone on the Team's Backlog

Pick two or three tasks from the team's backlog or the sample repository and write task briefs
Delegate to Codex, steer the work and deliver pull requests that pass the tests
Review with both Codex and teammates, then present what worked and what did not
Wrap up with a plan to bring Codex into the team and a starter set of AGENTS.md, a skill and config.toml

Schedule & training options

For individuals — public rounds

No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770

Join the waiting list Ask about rounds

For organisations — in-house / private

Tailor the content to your team’s tools and projects
Your dates, at your office or live online
Quotation with tax ID for procurement

Corporate training quote

Instructors

Samit KoyomFounder & CEO, IT Genius Engineering

Frequently asked questions

How does this course differ from AI Coding Agent for SDLC and Claude Code Roadmap?

AI Coding Agent for SDLC covers coding agents across the whole software lifecycle without tying to one tool, from planning, frontend and backend to testing and deployment. Claude Code Roadmap covers Claude Code in depth. This course focuses on Codex and how a team adopts it: AGENTS.md, skills, config.toml, pull requests, GitHub code review, Codex cloud, and sandbox and permission settings that follow company policy. It includes a comparison with Claude Code to help you choose. For Claude Code in depth, see Claude Code Roadmap; for the SDLC-wide view, see AI Coding Agent for SDLC.

Which accounts do I need, and does the institute provide them?

The institute does not provide accounts or licences. Learners use a personal or company ChatGPT account on a plan that includes Codex, and their own GitHub account. How much you can use Codex in a day depends on the plan. Please sign in to Codex and connect GitHub before the training day. Claude Code is demonstrated by the instructor in the comparison session, so you do not need an account for it.

Do I need to know how to program?

Yes. You should be fluent in at least one programming language and comfortable with Git, including branches, commits and pull requests, because you will read and decide whether to accept the diffs the agent produces. The course does not teach programming. If you are new to coding, learn a language and Git first.

Can I use my company's repository during the training?

Yes, if your company allows that code to be used with AI tools under the ChatGPT plan you have. Check the company policy before the training day, and make sure the repository holds no secrets or customer data. If you do not have permission, use the sample repository, which comes with tests and a backlog and covers every lab.

How do we use Codex safely in an organisation?

The course has a dedicated security session. It covers choosing the sandbox and approval mode for each kind of task, restricting network access, keeping .env files, tokens and customer data away from the agent, never running an agent with full access on production machines, reviewing every diff before a person approves the merge, and handling the licensing and intellectual property of AI-generated code in line with company policy.

Will the content stay current, given how often Codex changes?

Codex menus, modes and features change often, so the instructor updates the content and labs to match the latest version on the training day. The core practices, such as writing AGENTS.md, delegating and checking an agent's work, and setting permissions safely, still apply when the tool's interface changes.