Description
Build software with OpenAI Codex from written plans: configure AGENTS.md and agent settings, write verify-first acceptance checks, build in small tested pull requests, review with a second AI model and add security checks to CI.
Price$1,000Per person
Time10 course hoursOn-demand video and labs, plus 1 hour of coaching
DeliveryOn-demand asynchronous trainingWith online support and 1 hour of coaching
CredentialCertificate of completionFrom We Can Code IT
Your capstone, reviewed by a coach
You finish by taking an issue from a written plan to a merge whose checks pass, and your coach reviews the evidence with you: the plan, the tests and the second-model review. That review is part of earning your certificate.
Group coaching for teams
Each learner gets 1 hour of coaching. Teams pool their hours into group sessions, so a team of 6 shares 6 hours of group coaching.
About this course
This course teaches spec-driven development with OpenAI Codex, across the Codex CLI, the IDE extension and Codex cloud tasks. You set up the project instructions Codex loads every session, turn a specification into a plan whose every acceptance criterion has a verify command, and direct Codex to build it in small, test-first pull requests. Then you have Claude Code or another vendor's model review the same work adversarially, record decisions, and add security checks that run on every pull request. Shipped means the checks pass on the main branch, with evidence.
Who it is for
This course is for software developers, test automation engineers, technical builders, and entrepreneurs and founders who build their own products with AI coding agents.
Prerequisites
Introduction to GitHub with AI, or working knowledge of Git and pull requests
Comfort reading code in JavaScript or Python
Requirements
Works on Windows, macOS or Linux
Git, Node.js or Python, and a code editor such as Visual Studio Code (all free), and a free GitHub account
Labs use a public practice repository, since branch protection requires a paid plan for private repositories; do not put employer code in it. Teams on a paid GitHub plan can use a private repository.
Codex through a paid ChatGPT plan (Plus or higher) or a paid OpenAI API key; the free plan does not include the Codex cloud tasks used in the labs
A second reviewer from a different vendor, such as Claude Code, which needs a paid Claude plan (Pro or higher)
What you will be able to do
- You will be able to install and configure OpenAI Codex, including AGENTS.md project instructions, approval modes and sandbox settings.
- You will be able to write a plan file whose acceptance criteria each carry a verify command that passes when built and fails when not.
- You will be able to direct Codex to build from a plan in pull requests under 400 lines, with a failing test written before each fix.
- You will be able to run an adversarial code review with a second AI model from a different vendor, make the reviewers argue, and act on the findings that survive.
- You will be able to configure a CI pipeline that runs tests and security checks for access rules and endpoints on every pull request, with Codex running unattended where useful.
- You will be able to record architecture decisions and risks so later engineers and agents understand why the code is the way it is.
Syllabus
1
Setting up OpenAI Codex 1 hour
The Codex CLI, IDE extension and cloud tasks, Approval modes, sandbox settings and config.toml, What agents do well and badly, Agents open pull requests and never merge them.
Lab: You will install Codex, restrict what it may do, and compare its output on one task with and without a specification.
2
Project instructions and coding standards 1.5 hours
AGENTS.md project instructions, Coding defaults and owned paths, Skills, custom prompts and MCP servers, Rules in files, not in memory.
Lab: You will write AGENTS.md for a repository, add one reusable command, and confirm Codex follows them on a test task.
3
Plan files and verify-first acceptance checks 1.5 hours
Exploring the codebase before building, Plan file structure, "When X, the system shall Y" criteria, Verify commands that fail when the feature is absent.
Lab: You will have Codex explore the code and turn a product owner's issue into a plan file with a verify command for every criterion, and prove each fails before the build.
4
Building in small, test-first pull requests 1.5 hours
Pull requests under 400 lines, Failing test before the fix, Baseline test runs, Labeling agent-authored work.
Lab: You will have Codex build one plan item as a pull request under 400 lines, with a failing test first.
5
Cross-model adversarial review 1.5 hours
Codex's own review commands, Claude Code or another vendor's model as second reviewer, Review prompts that demand vulnerabilities and refusals, Making reviewers argue and triaging findings.
Lab: You will review your pull request with Codex and with Claude Code, make the two debate their findings, and fix what survives.
6
Automation, security checks and decision records 1 hour
codex exec and Codex code review on GitHub, Access-rule and tenant isolation tests, Required checks, Architecture decision records and a risk register.
Lab: You will add a security check and an automated Codex review step to CI, and write a decision record for one design choice.
7
Capstone 2 hours
Project instructions, Plan with verify commands, Codex build, Cross-model review, Verified merge.
Lab: You will take an issue from plan to a merge whose checks pass on main, and walk your coach through the evidence.
+
Coaching sessions 1 hour
Your coach reviews your capstone with you and answers questions as you work. Teams pool their hours into group sessions, so a team of 6 shares 6 hours of group coaching.
How it is delivered
How you learn
On-demand asynchronous training with online support, plus 1 hour of coaching.
Live or in person
Live online or in-person sessions are available for an additional fee. Email admissions@wecancodeit.org for details.
Frequently asked questions
What is spec-driven development with Codex?
You write a plan with acceptance criteria and a runnable check for each before Codex writes code. Codex builds to the plan, and the checks prove whether it worked.
Spec-driven development vs vibe coding: what is the difference?
Vibe coding prompts an AI until something works. Spec-driven development writes the target and its checks first, so the result can be verified, reviewed and maintained.
Why review with a second AI model?
A model reviewing its own work tends to repeat its own reasoning. A model from another vendor catches different problems, and making the two argue surfaces the findings worth fixing.
Is there a version of this course for Claude Code?
Yes. Spec-Driven Software Development with Claude Code covers the same method and hours with Claude Code as the main tool. Choose the tool your team already uses.
What does the coaching include?
Each learner gets 1 hour of coaching. Your coach reviews your capstone with you and answers questions as you work through the labs. Teams pool their hours into group sessions, so a team of 6 shares 6 hours of group coaching.
How long does it take and what does it cost?
10 course hours of on-demand video and labs plus 1 hour of coaching, for $1,000 per person. A paid ChatGPT plan or OpenAI API key for Codex is extra.
Related courses
Build software with Claude Code from written plans: configure CLAUDE.md and agent settings, write verify-first acceptance checks, build in small tested pull requests, review with a second AI model and add security checks to CI.
10 course hours, $1,000
Set up and use GitHub for real work: repositories, branches, pull requests, issues, Projects, Actions, Copilot and Codespaces, security settings, with preparation for the GitHub Foundations (GH-900) exam.
8 course hours, $800
Test, attack and secure an AI-generated web application with automated end-to-end tests, multi-model AI code review and hands-on red-teaming of access control, injection, secrets and AI features.
10 course hours, $1,000
Write specifications and executable acceptance criteria that AI coding agents can build from, configure GitHub issue forms and Projects to manage them, and verify delivered work in deploy previews.
10 course hours, $1,000
Set up a shared delivery system for a product owner and engineer working with AI coding agents: a working agreement, a weekly cycle, two approval gates in GitHub, and evidence-based reviews.
8 course hours, $800
Enroll your team
One or a few learners
Choose the number of seats, add the course to your cart and check out.
A group, or a written quote
Send the form below for group enrollment or a written quote. We reply within one business day. You can also email admissions@wecancodeit.org.
We Can Code IT, LLC, 26300 Cedar Rd, Suite 1105, Beachwood, OH 44122. Registered with the Ohio State Board of Career Colleges and Schools, Registration No. 15-03-2064T.
