OpenAI Codex for Everyone

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

The Finished Workflow (8 minutes)

Cold open and live demo Open cold on a finished automation, a Monday morning brief that read a folder of project files, summarized what changed, and drafted a status update on its own The contract for the day, you will see exactly how this works and build its core yourself What task does your team repeat every week that nobody wants to do?

Part 1: From Chatbots to Agents (38 minutes)

What agentic workflows are, and the harness tour

• The arc we all lived, ChatGPT as pure chat, then images, then tools and search, and now an AI that does the work itself, in loops, until the task is done
• Chat gives you answers, agents give you finished work, the difference is the loop and the harness
• One agent, three surfaces, the Codex App where we live today, the IDE, and the CLI, same brain and same settings everywhere
• The tools at its disposal, reading and writing files, running commands, searching, and browsing your project
• The agent loop, prompt, think, use tools, repeat until done, and context as the agent’s working memory

Models and effort—your control panel

• The agent’s brain is swappable, Sol for deep reasoning, Terra as the balanced default, Luna for fast and cheap work
• Reasoning effort from low to max, how hard it thinks before acting, and what that costs
• A decision rule for normal humans, the balanced default for daily work, deep reasoning for ambiguous or high-stakes tasks, fast and cheap for quick mechanical work

Hands-on—your first delegation

• Everyone opens the sample project folder in the Codex App
• The four-part prompt, goal, context, constraints, and done when
• Voice dictation as the unlock, talk to it like a colleague instead of composing essays
• First task, summarize what is in this folder and flag what is out of date
• The safety net in one breath, by default Codex only touches this workspace and asks before anything bigger

Plan mode—make it interview you

• Without Plan mode Codex has a bias for action, with Plan mode it touches nothing until you approve
• Watch Codex ask clarifying questions and turn a fuzzy request into a concrete plan
• The plan is an artifact, iterate on it, save it, hand it to a teammate

Part 2: Teach It How You Work (32 minutes)

Guidance files—what makes a good one

• The repeated-instructions problem, if you keep typing the same guidance, stop
• A guidance file is a playbook the agent reads automatically at the start of every session
• The anatomy of a good one, what the project is, how things run, conventions, do-not rules, and what done means
• The retrospective loop, tell Codex to update its own guidance file so a mistake never happens again

Connectors—plug in to the tools you already use

• Your context lives in your docs, tickets, and drives, not in local files, and connectors let Codex pull live data from those systems
• Demo, connect one real service and watch Codex pull live information into a deliverable
• The restraint rule, start with one or two connectors that remove a manual loop you already do often

Hands-on—review like a pro

• Codex does the work, you do the judgment, reading the change view and sending feedback on specific rows
• The built-in review workflow as your quality gate, even on documents
• Run a real change through review on your own sample project
• Q&A

Part 3: Multiply Yourself (28 minutes)

Skills—package what works

• The handoff problem, good workflows die on a single laptop
• A Skill in plain terms, a repeatable task packaged with instructions and templates so anyone, or any agent, can run it by name
• Demo, turn the morning brief workflow from the cold open into a named Skill

Hands-on—build your own skill

• Build one skill from your own recurring task, the one you named in the opening poll
• Design rules, a sharp description with trigger phrases, clear inputs and outputs, one job per skill
• Where skills live, personal versus shared with your team

Automations—work while you sleep

• Scheduled tasks, pick the project, the prompt, and the cadence, results land in your inbox
• Skills define the method, schedules define the rhythm, only automate what already works reliably by hand
• The self-improving Skill, an automation that reviews its own recent runs and sharpens its own instructions

Audience Choice (8 minutes)

• Audience choice—the class votes on a workflow to run live or an earlier topic to go deeper on
• Overflow room for questions and hands-on catch-up

Your 30-day plan and resources (6 minutes)

• Week 1: daily delegations with the four-part prompt, and try the same task at two effort levels to feel the difference
• Week 2: write your guidance file and connect one tool
• Week 3: package your best workflow as a Skill
• Week 4: schedule it and share it with a teammate
• Going further: parallel work with multiple threads and worktrees once the basics stick
• Resources and community
• Q&A