By the end of this lesson you will have:
The inbox is agent-run. The calendar is agent-run. One job is left: production in Bopgpt itself. This lesson points your agent at Bopgpt so it can set up and run the same production work you learned to do by hand.
Everything agents did in the last three lessons ran through official connections. This lesson goes further: modern AI assistants can operate software directly. They can open a browser tab, see the screen, and click the buttons. That means the same assistant that plans her calendar can sit in Bopgpt and execute the plan. The team behind this course runs parts of its own operation this way.
This is also the newest and least polished layer of the stack, which is why it is taught last and always supervised. Expect this lesson to update faster than any other as the official integrations land.
One rule governs the whole lesson, and it has a reason. Only delegate a job you have already mastered by hand. The agent copies your process, including its flaws. Phase 3, starting at lesson 3-3 (The Content Factory), made you good at production; that is what makes handing it over safe and fast instead of a faster way to waste credits.
The three routes for agent-driving Bopgpt, ranked by what works today:
These are hard rules for any agent that touches production:
In your AI assistant: check whether it can operate a browser, through an extension or a computer-use mode. If it can, you are on route 1, which is where most students start with the course-default assistant. If it cannot, run the planning half of production through the Arsenal and keep clicking Bopgpt yourself until route 2 ships.
In your AI assistant: give it the job the way you would brief a trainee.
You're operating Bopgpt in the browser tab I've opened. Today's
brief: [scene, outfit count, keepers needed]. Her settings: [paste]. Rules:
stay inside her content ceiling, never touch the never list, generate only β
nothing publishes without me. Restate the job, then begin, narrating each step.
When the agent restates the job correctly before touching anything, it is briefed.
In Bopgpt, with the tab open and the agent driving:
You will not save time on the first run. You are training the process, and the second run is where the time comes back. When the batch is finished, agent-executed and human-approved, you know exactly where it still needs you.
In your AI assistant: decide what this agent has earned, write it as a job description, and save it in the Arsenal with its rules, the same as every other prompt. For most operators the first delegated job is one of these:
When the job is saved, the section's work is complete: inbox, calendar, numbers, and production all have a standing agent job.
~20 minutes, supervised.
An agent has produced a batch in Bopgpt under your supervision, and its standing job is saved in the Arsenal. Mark the lesson complete.
Normal at this stage of the tooling. Keep the Bopgpt tab focused, keep jobs small (one shoot, not a day's production), and narrate corrections. If your assistant is simply not good at browser work yet, stop forcing it: run the planning half through the Arsenal and keep clicking yourself until route 2 ships.
It is your account, your machine, and your supervision, and that is the model this lesson teaches. What it never does: run unattended, publish anything, or touch fan-facing surfaces. Those rules are what makes this safe to run today.
The official Bopgpt integration decision is in motion. When it lands, this lesson updates and the course announces it. Build the supervised-run habit now. The same habit runs on the official connection the day it ships.
That closes the AI Advanced section. Next is the Vault, starting with lesson v-1 (Scene Swap): the optional expansion plays, and the operating week in lesson 6-5 (The Operating Week) is the long-game routine you run from now on.