In this hands-on workshop participants will experience an idea-to-app journey powered by three complementary AI helpers — ChatGPT (O3), Codex CLI, and Aider. By the end, every learner will have a fully-functioning Flask “To-Do List” and a clear mental model of when to reach for each tool, how to steer it, and how to keep themselves “in the loop.”

1. Workshop at-a-Glance

Item	Detail
Duration	~90 min (can stretch to half-day with extensions)
Format	Live coding, short sprints + debriefs
Learning Goals	1) Draft specs with an LLM, 2) Scaffold code autonomously with Codex CLI, 3) Iterate & refactor with Aider, 4) Apply git-first review habits
Prerequisites	Laptop with Python 3.11+, git, an OpenAI API key, Node/conda optional
Install Beforehand	pip install codex-cli aider (OpenAI Codex CLI - GitHub, GitHub - Aider-AI/aider: aider is AI pair programming in your terminal)

---

2. Environment Setup (15 min)

1. Clone starter repo – an empty todo-workshop directory with .gitignore.
2. Configure keys: export OPENAI_API_KEY=….
3. Smoke-test tools

   codex --version    # should print model list
   aider --help       # should show CLI flags

Facilitator tip: keep a USB stick or codespace ready for anyone with install issues. Preparing a sandbox in advance matches best-practice workshop design guidance on removing friction points. (The Design of a Workshop: A Guide to Designing Workshops that Engage)

---

3. Activity Flow

Step 0 – “Blueprint in 5 minutes” (ChatGPT O3)

Goal: produce a concise spec.
Prompt:

“Draft an architecture for a one-page Flask to-do app: key routes, data model, and files.”

Checkpoint: learners paste the generated bullet list into spec.md. O3’s high-level reasoning is well-suited for this ideation phase (Introducing OpenAI o3 and o4-mini).

---

Step 1 – “Scaffold at Mach 1” (Codex CLI) 20 min

Action	Command	What to Highlight
Launch	codex --approval-mode full-auto "Create a Flask to-do list app..."	Full-auto lets the agent run, yet commits every milestone to git for audit. (codex/codex-cli/examples/prompting_guide.md at main - GitHub, OpenAI Codex CLI – Getting Started - OpenAI Help Center)
Observe	watch terminal log	Learners see files (app.py, templates/index.html, requirements.txt) materialise.
Inspect	git log --oneline	Stress habit of reviewing each AI commit. Git integration is a first-class feature. (codex/codex-cli/examples/prompting_guide.md at main - GitHub)

Talking point: Codex CLI is a terminal-native agent that can run code, install deps, and iterate while keeping your repo local. (OpenAI Codex CLI - GitHub)

---

Step 2 – “Iterate Like a Senior Dev” (Aider) 35 min

1. Open chat on scoped files

   aider app.py templates/index.html

2. Feature request:

   “Add ability to mark tasks complete, strike them through, and clear completed items.”

3. Review proposed diff; accept; watch auto-commit messages such as “feat: complete/clear tasks”. Aider’s multi-file edits plus repo map give GPT-4 context across the project. (GPT code editing benchmarks - aider, Git integration - aider)

4. Optional refactor sprint: ask Aider to prevent empty tasks or add Bootstrap styling.

Facilitator tip: encourage participants to reject at least one AI diff to practise control. Good workshops build agency, not automation dependency. (The Design of a Workshop: A Guide to Designing Workshops that Engage)

---

Step 3 – “Play & Test” (Manual) 15 min

• Run python app.py; open localhost:5000.
• Exercise add/complete/clear.
• If a bug appears, jump back into Aider with a natural-language fix request. This live debugging mirrors the “complex change” example in Aider docs. (A complex multi-file change, with debugging - aider)

---

4. Reflection & Discussion (10 min)

• Which tool felt most magical?
• Where did human judgment matter?
• How would you productionise this? (tests, DB, CICD).
  Use research on workshop debriefs to cement learning by linking emotion → insight → action. (The Design of a Workshop: A Guide to Designing Workshops that Engage)

---

5. Stretch Goals (if time)

Idea	Tooling Pointers
Add SQLite persistence	Ask Aider to integrate sqlite3; reference Flask todo tutorials for schema pattern. (To-do List App Using Flask (CRUD Features) - Pythonista Planet, [Todo list app using Flask
Dockerise app	Let Codex CLI author Dockerfile; learners review layers.
API tests	Generate pytest suite with Aider; run pytest -q.

---

6. Facilitator Checklist

• Prep: Confirm internet access (for OpenAI calls) or provide proxy endpoint.
• Safety nets: Use a branch per attendee; enforce --sandbox if machines allow.
• Time boxes: 5 min buffer after each coding block for lagging laptops.
• Support channels: Slack/Teams thread for copy-paste commands.

---

7. Reference Links

• Codex CLI repo (OpenAI Codex CLI - GitHub)
• “Getting Started with Codex CLI” guide (OpenAI Codex CLI – Getting Started - OpenAI Help Center)
• Codex prompting & autonomy modes (codex/codex-cli/examples/prompting_guide.md at main - GitHub)
• Aider repo & features (GitHub - Aider-AI/aider: aider is AI pair programming in your terminal)
• Aider git integration docs (Git integration - aider)
• Multi-file editing benchmarks (GPT code editing benchmarks - aider)
• Complex change + debug example (A complex multi-file change, with debugging - aider)
• Workshop design best practices (The Design of a Workshop: A Guide to Designing Workshops that Engage)
• Flask To-Do tutorial for reference (To-do List App Using Flask (CRUD Features) - Pythonista Planet, Todo list app using Flask | Python - GeeksforGeeks)
• OpenAI O3 model capabilities (Introducing OpenAI o3 and o4-mini)

---

Outcome

Participants leave with:

• a running app they can push to GitHub or extend later,
• a repeatable workflow for small-to-medium tasks: Plan → Scaffold → Iterate → Test, and
• confidence guiding AI while retaining software craftsmanship.