A university-level guide summary for non-majors and planners: set up a project Workspace with Codex Client and develop specs for teamwork.
A regular ChatGPT web chat and the project-based, multi-file Codex Client take different approaches to designing requirements.
| Compared | Regular ChatGPT (web browser) | Codex Client (Workspace-based) |
|---|---|---|
| What it sees | Only the one-off text typed in the input box | Every file in the project folder (Workspace) |
| How files are handled | Copy and paste result code from the screen by hand | AI creates and partly edits the chosen markdown files directly |
| Multi-file reasoning | You must upload earlier files by hand every time to update linked content | When one plan changes, it reasons about linked files at the same time |
We set a dedicated folder on the desktop, automation, as the AI Workspace and practice there.
Under this folder structure, AI fully links the input-output chain of each spec file, from the step 0 summary notes to the step 5 email report.
Starting from a long, raw meeting transcript, you'll produce a final approved spec with lint errors fixed, and even send a decision request email.
requirements-specification.md: combined business requirements and flow docClick each step to update the card below with how data is transformed (Input -> AI -> Output).
Create a dedicated local folder and open that path in Codex AI or your editor, so there's a channel to read and write markdown spec docs.
Click the buttons on the left in order to see **where to act (red guide box)** — where a learner actually clicks and types in Codex Client.
Connect the empty automation folder on the desktop.
Use a prompt to create meeting-notes-raw.md.
Tell it which file to read and which to write for the summary
Press the build button on the right to create the summary.
Double-click the created meeting-notes.md in Explorer.
Enter a prompt and it's applied to files in the project folder.
A logical architecture map of how AI sees files in the project folder and carries planning forward step by step.
The markdown specs produced at each step don't stop at planning. In the Day 2 ~ Day 10 course, they carry forward as the key base data (Context) Codex uses to automatically write DB ERD designs, back-end API specs and front-end wireframe code.
During development, the CEO added a new requirement at the end of the meeting notes.
"Let's add account deletion too. When a user deletes their account, all booked counseling sessions must be cancelled automatically and refunded right away."
meeting-notes-raw.md."Using the new conversation in meeting-notes-raw.md, update functional-requirements.md with the account deletion feature (FR-005) and the booking cancellation exception flow that follows."We add 3 Tasks to the Day 1 curriculum to clearly link technical skills and AI teamwork skills. The goals and outputs of the existing Tasks/Missions stay the same; the extra Tasks go deeper in designing, building and checking those results.
| Task ID | Extra Task | Area | Tech Skill | AI Skill |
|---|---|---|---|---|
ADD-D01-T01 |
Define the service problem and Personas | HYBRID | TECH-BA TECH-DDD |
AI-CTX AI-ANA AI-DOC |
ADD-D01-T02 |
Sort Features into Domains | TECH | TECH-BA TECH-DDD |
AI-ANA AI-REV |
ADD-D01-T03 |
Review the Domain Map and responsibility boundaries | AI | TECH-DDD TECH-ARCH |
AI-CTX AI-REV AI-DOC |
ADD-D01-T01 Define the service problem and Personas → ADD-D01-T02 Sort Features into Domains → ADD-D01-T03 Review the Domain Map and responsibility boundaries
01_SERVICE_OVERVIEW.md, 03_DOMAIN_MAP.md