About This Course
<p>This course contains the use of artificial intelligence. <span>Coding has changed.</span> AI agents now write a large share of new code. Google's CEO says 75% of the company's new code is AI-generated. Writing code is no longer the hard part. Judging it is: knowing what to ask for, checking what comes back, and shipping it safely.</p><p><span>Vibe coding breaks in production.</span> Describe an app, accept whatever comes back, ship it. It feels fast, and the reports show what it ships:</p><ul><li><p>Veracode tested more than 100 large language models on 80 coding tasks. The code had security vulnerabilities in 45 percent of cases.</p></li><li><p>Escape scanned 1.4K vibe-coded apps and found 2,038 highly critical vulnerabilities, 400+ leaked secrets and 175 instances of exposed personal data.</p></li><li><p>A researcher found 170 of 1,645 apps built on Lovable with database tables open to anyone (the vendor disputes the CVE).</p></li><li><p>On Replit, a coding agent deleted a founder's production database during a code freeze, by his own account.</p></li></ul><p>No model fixes this by itself, and Codex won't write secure code on its own either. What was missing is the engineering around the agent.</p><p><span>This course is software engineering with OpenAI Codex.</span> You take one real product, <span>NoUploadPDF</span>, through the whole software development life cycle (SDLC), from requirements to a live v1.0 release, working in the Codex app.</p><p><span>What you build</span></p><ul><li><p><span>NoUploadPDF</span>, a real PDF toolbox with 11 tools (merge, split, extract pages, rotate, reorder and delete, page numbers, watermark, images to PDF, PDF to images, compress, and one merge-number-compress chain), all running in your browser. Your files never leave your computer.</p></li><li><p><span>The in-browser engine:</span> a tool registry, one shared tool page and a Web Worker that runs every job with pdf.js and pdf-lib.</p></li><li><p><span>A CI/CD pipeline:</span> every change goes through a pull request, GitHub Actions runs unit and end-to-end tests, Vercel builds a preview, and only green checks reach production.</p></li><li><p><span>v1.0, tagged, released and live</span> on Vercel.</p></li></ul><p><span>What you learn, section by section</span></p><ol><li><p><span>Getting started:</span> what Codex is (app, CLI, IDE extension, cloud), plans and usage limits, models and reasoning effort, install on Mac or Windows, Node.js, Git, VS Code, config.toml and where the Codex home folder lives.</p></li><li><p><span>System design:</span> functional and non-functional requirements, an architecture decision record (ADR) for client-side vs server-side processing, high-level architecture and data flow, the SDLC and the CI/CD pipeline.</p></li><li><p><span>Codex app fundamentals:</span> projects and chats, the composer (/ commands, @ mentions, $ skills), <span>/plan</span> and Plan mode, a Next.js scaffold, steering and queuing, reviewing diffs, the app's terminal, shortcuts.</p></li><li><p><span>CI/CD pipeline:</span> Git and GitHub, SSH keys, continuous integration with GitHub Actions, continuous deployment on Vercel, branch → pull request → preview → production, rollback.</p></li><li><p><span>Configuring Codex:</span> <span>/init</span> and the AGENTS file, memories, settings vs config.toml, spec-driven development (PRD, ADRs, tasks), context engineering with <span>/status</span> and <span>/compact</span>.</p></li><li><p><span>Security guardrails:</span> the sandbox and approval policies, permission modes, auto review, a permission profile that keeps .env secrets out of reach, rules for risky commands, trusted projects and web search.</p></li><li><p><span>Core architecture, MCP and skills:</span> the shared shell and engine, <span>/mcp</span> with the Context7 MCP server for current library docs, Codex skills and $skill-creator.</p></li><li><p><span>Automated testing:</span> the test pyramid, unit tests with Vitest, end-to-end (E2E) tests with Playwright in CI, branch protection, test-driven development (TDD).</p></li><li><p><span>Feature development I:</span> business logic and edge cases, parallel work with Git worktrees, a DESIGN file and image input, user acceptance testing (UAT) in the built-in browser, hooks as quality gates, subagents and custom agents.</p></li><li><p><span>Feature development II:</span> conversion in the browser, <span>/goal</span>, PDF compression and licences, scalability, threat modeling, security scanning with the Codex Security plugin.</p></li><li><p><span>Debugging and production readiness:</span> root cause analysis, browser debugging, technical debt, architecture documentation, <span>/review</span> code review, Lighthouse and Core Web Vitals, a scheduled nightly smoke test.</p></li><li><p><span>Extending Codex:</span> plugins, connectors (GitHub, the Vercel plugin and MCP), @Chrome and Computer Use, codex exec and the SDK.</p></li><li><p><span>The Codex CLI:</span> the same agent in a terminal: start a session and read its screen, follow-ups, steering and queuing, images, the sandbox and approval flags with <span>codex sandbox</span>, <span>/model</span>, <span>-m</span> and <span>-c</span> overrides, web search, <span>codex resume</span> and <span>codex fork</span>, worktrees, <span>codex review</span>, and <span>codex exec</span> for scripts.</p></li><li><p><span>Prompt injection and cost:</span> how prompt injection tricks an agent, a live demo with a PDF that talks to Codex, cost optimization with /status and /compact, moving your setup from Claude Code or Cursor.</p></li><li><p><span>Local LLMs (optional, Windows):</span> model providers, a local model with Ollama, a local model with LM Studio.</p></li><li><p><span>Scaling and the capstone:</span> server-side architecture, production hosting, <span>/plan</span> for an upgrade written as GitHub issues, pipeline design, and the final build with every feature, shipped as the v1.0 release.</p></li><li><p><span>Bonus:</span> what to learn next, how to stay in touch, and how to get and share your certificate of completion.</p></li></ol><p><span>Cheat sheets for every section.</span> Sections 1–16 each come with a PDF holding every prompt, command and shortcut from that section, attached to its first lecture.</p><p><span>Mac and Windows.</span> The course is recorded on a Mac. Where Windows really works differently (installing the app, the sandbox, finding the Codex home folder, the optional local-model section), a Windows lecture shows it.</p><p><span>Real, not staged.</span> Every screen is the real Codex app, GitHub, Vercel and the live site. The product's full code is public on GitHub, so you can compare your project with mine at any time.</p><p><span>Quizzes.</span> Five sections end with a 5-question quiz (System Design, CI/CD Pipeline, Security Guardrails, Automated Testing, and Prompt Injection, Cost Optimization and Moving from Claude Code or Cursor), so you can check what you learned.</p>
What you'll learn:
- Build and ship a real web app with OpenAI Codex through the whole SDLC, from requirements to a live v1.0 release
- Turn an idea into requirements, an architecture decision record (ADR) and a data-flow design before Codex writes code
- Use Codex slash commands in real work: /plan, /init, /status, /compact, /mcp, /goal and /review
- Set up a CI/CD pipeline: Git, pull requests, GitHub Actions checks, Vercel preview deployments, production and rollback
- Make tests Codex's feedback loop: unit tests with Vitest, end-to-end (E2E) tests with Playwright in CI, and TDD
- Keep an AI coding agent safe: sandbox, approval modes, a permission profile that hides secrets, rules for risky commands
- Configure Codex with an AGENTS file, config.toml and a written spec, and manage its context and usage
- Extend Codex with MCP servers, skills, hooks, subagents, worktrees, plugins, connectors and codex exec
- Spot prompt injection and threat-model a web app, then security-scan it with the Codex Security plugin
- Debug with root cause analysis, review code with /review, and get a site production-ready with Lighthouse