The Third Brain · Coming soon
Stop watching Claude Code work. Wake up to drafts awaiting review
This course sets Claude Code to work through a list of tasks while you are away, one task per run, inside limits on time and scope that you set. Finished work waits in a review queue with a log of what happened. Nothing it produces is accepted until you read it and say yes.
The course: Agents that run while you sleep: loops, schedules and review gates
$19 $39 early-bird price
This course is being built now. Join the early-access list to reserve the early-bird price and shape what it covers.
Joining the list is free. You only pay if you buy at launch, at the early-bird price.
If this sounds like your week
- You sit and watch Claude Code work because you do not trust it alone.
- On a long run you start clicking approve without reading what you are approving.
- You have a backlog of similar tasks that only moves while you are at the keyboard.
- You have read the stories of agents deleting files, and skip-permissions mode feels like a gamble.
What you do now, and where it stops working
Today you either babysit every run or run the same jobs by hand each week. The tempting shortcut is to switch off the permission prompts and hope, which trades your time for a risk you cannot see until morning.
What changes after the course
- A repeat job runs at the time you set and leaves its output in the same place every time.
- A loop works through a task list one item at a time and stops cleanly when the list is empty.
- In the morning you open one review queue and accept, edit or reject each result.
- A run log tells you what each run did, so a bad run is something you read about, not something you clean up.
How it works: Queue, limits, gate
- 1Queue: write the work as a task file, one small and checkable task per line.
- 2Limits: give every run one task, a time limit, a folder it may write to and a condition that stops it.
- 3Gate: runs produce drafts in a review queue; nothing is merged, sent or published without your approval.
- 4Log: each run writes what it did and why it stopped, so you can check the night in a few minutes.
What you get
- Schedule and loop templates for Claude Code
- Task queue file format with examples
- Review gate checklist
- Run log template for checking what happened overnight
- Video walkthrough of an overnight run from setup to review
This is for you if
- You already use Claude Code for real tasks and spend time watching it run.
- You have jobs you repeat every day or week, or a long list of similar tasks.
- You want unattended work only where you can review every result before it counts.
Skip it if
- You have not used Claude Code yet; start with the memory course or a Claude Code introduction first.
- You want agents to publish, send or deploy on their own with no human review.
- Your work needs a person to make a judgement call at each step.
Before you join
- What if it deletes or breaks something while I am asleep?
- Each run may write only to the folder you give it, and its output lands as a draft in a review queue. Nothing changes your real files, sends or publishes until you approve it in the morning. Jay West runs his own scheduled loops behind the same kind of review gate.
- What if it runs up a bill overnight?
- Every run is one task with a time limit and a stop condition, so a run cannot keep going all night. The course has you start with a single scheduled job and check its usage before you add more.
- Which jobs are safe to hand over?
- Jobs with a clear finish and an output you can check quickly: drafts, summaries, first passes through a backlog. The course gives you a test for each candidate job before you schedule it.
- Is this just skip-permissions mode with extra steps?
- No. Turning off the prompts removes your review. This setup moves your review to the morning, where you can read each result calmly, instead of approving steps one by one at the keyboard.
- Do I need a server or to be a developer?
- No. The course runs on your own computer, using a schedule you set there, and every file it uses is plain text you can read.
Rather have it built for you? Describe the project and the same work can be done for your team.