Choose the desktop app
Students can install Claude Desktop and use the Claude Code tab, or download Codex. Jay currently recommends Codex because many students have run into Claude age restrictions.
Level 1
Start with an agentic harness.
Level 1 ends when a student has a desktop AI app installed, a starter harness in the right file, and a first project folder that can survive more than one AI session.
Alfred is the starter version.
Jay's starter harness is named Alfred after Batman's butler: the operator behind the scenes who tracks what matters, prepares the tools, keeps the system organized, and helps the hero act with more focus.
The student is still the hero. Alfred is a first assistant, not the whole curriculum and not the student's replacement.
Setup checklist
The first session should produce infrastructure.
A student who only chats with AI starts cold every time. A student with a harness starts with project memory, files, rules, verification, and a path to the next task.
Create a project folder
Use one folder per serious project. The folder is where the harness stores continuity, data, outputs, and working notes.
Add the harness file
Use AGENTS.md for Codex or CLAUDE.md for Claude Code. Download Jay's Codex starter file, place it in the project folder, and let Codex read it before the first project prompt.
Create SESSION.md and MEMORY.md
These files turn one AI session into a continuing project system.
Run a first verified task
Ask the harness to choose a starter task, define success criteria, complete it, and record what changed.
Starter folder
The project operating system.
AGENTS.md
The Codex harness: role, team, rules, integrity constraints, project structure, and verification standards.
SESSION.md
The current state of the project, open loops, next actions, and what was verified last time.
MEMORY.md
Append-only lessons from mistakes, repeated friction, and workflow improvements.
output/
The place where finished proof goes: reports, project plans, deployed-site notes, and submission packets.
First prompt
Give Codex a job the harness can manage.
The first prompt should not ask for magic. It should ask Codex to inspect the folder, explain the harness, and create a first project sequence.
Read AGENTS.md and introduce yourself as Alfred. Then help me set up this project folder. My current project idea is: [write the idea here] I want you to: 1. identify what files this project needs, 2. create a first SESSION.md, 3. create a first MEMORY.md, 4. propose the first five tasks, 5. explain what you will verify before calling any task done.
Claude reference
Claude remains a reference path.
The workshop starts with Codex because student access to Claude can be age-restricted. The Claude harness still matters as a reference, especially for students who later use both Codex and Claude for different kinds of work.
The model landscape changes. The durable skill is not loyalty to one app. The durable skill is learning which model should plan, which model should build, which model should review, and which work can move to local compute.