DAY 01 / SELECT
Find the leverage
Choose one recurring task that is small enough to improve and safe enough to review.
Why this matters
A useful first agent starts with a real job, not a model comparison. Look for work that happens often, follows a pattern, and leaves an output a person can judge.
A useful first agent starts with a real job, not a model comparison. Look for work that happens often, follows a pattern, and leaves an output a person can judge.
Try these steps
- List three tasks from the last week that repeated, waited on you, or moved between tools.
- Score each task from 1 to 5 for frequency, time cost, repeatability, and verifiability.
- Reject any task where a mistake is hard to notice or could send, spend, publish, or change something without approval.
- Shrink the best candidate to one trigger and one output. Write a one-page system brief.
Illustrative example
A consulting team has five client calls a week. Turning approved notes into a recap and draft follow-up takes 20 minutes. It repeats, the inputs are available, and the meeting owner already checks the email before sending. That is a better first project than automating all client communication.
Watch for: Do not pick a high-stakes decision, an occasional task, or work where you cannot tell a good result from a bad one.
The member lesson adds an editable workbench. Draft your output, check your exit criteria, and download your work. Progress is saved to your member account, with a backup on this browser.