Directives
Durable rules, sources, boundaries, and checks define what strong work means.
For owner-operators who have real work to do
Install Codex, connect the tools where your work already lives, and turn one repetitive task into a system you can use next week.
YOUR BUSINESS / YOUR MACHINE
What to expect
You will build on your own laptop, with your own business in mind. The point is not an impressive chat. The point is a useful file and a process you can run again.
The room, in four moves
Get the ChatGPT desktop app open and reach Codex on your machine.
Open a clean starter built around directives, orchestration, and execution.
Connect the actual source system and create a useful, reviewable output.
Correct one failure, define the next test, and schedule a seven-day trial.
The operating model
Durable rules, sources, boundaries, and checks define what strong work means.
The Codex agent reads the job, finds the right context, routes the work, and verifies the result.
Small deterministic tools handle mechanical steps that should work the same way every time.
Before you arrive
Do the first three before the lab. Arrive ready to connect the system that holds the real work. If you hit an install or account-access problem, tell Hector before the session.
Mac or Windows is fine. You will work on your own machine.
Make sure you can sign in at chatgpt.com before you arrive.
If your laptop is managed by work, check your install permissions.
Pick something annoying, frequent, and easy to recognize when done well.
Be ready to sign in to the Gmail, CRM, Drive, or other system the workflow actually uses.
Use live systems with guardrails
We will connect only the accounts the workflow needs and start read-only where possible. Sending, publishing, deleting, or changing records stays behind explicit approval. You handle your own login, and passwords or access keys never belong in a prompt or workshop file.
Good first workflows
What you leave with
Codex running on your laptop
A clean workspace built around your business
The real source system connected with narrow permissions
Rules that tell the agent what good work looks like
One tested workflow that creates a real file
A seven-day plan to prove whether it is worth keeping
Quick answers
No. You need to know your business task and what a good result looks like.
You can, but you do not have to. Installation is part of the lab. Do test your ChatGPT sign-in and confirm that your laptop allows app installs.
Tell Hector before the lab so there is time to plan a fallback.
Yes. That is how the workflow becomes useful. We use only what the job needs, begin read-only where possible, and confirm before anything is sent or changed.
No. You sign in on your own machine and manage the authorization yourself. Never paste a password or access key into Codex.
We can use an export or local file to prove the workflow first, then finish the connection after the useful process is clear.
Your two-minute prework