AI cost consultant, 6 years at Amazon

One live Saturday, 6 hours. You already use Claude Code or Codex every day. In this workshop you set them up to work when you are away from the keyboard.
We build it together in your own repo and on your own machine. At 4:45 pm ET your own loop runs on your machine, with a cost gate and a review sheet. It keeps running after class.
I run this setup every day. My loop ran 2,328 unattended jobs between Nov 7 and Nov 10, 2026, and 2,306 of them finished without an error. I also made the mistakes you will make. One script burned 1.63 million tokens ($8.92) on work my session had already done. Bedrock calls cost me $47.53 before I added a gate. You get the gates I built after each mistake.
You get the recording and the repo. Each class has a limit of 25 seats.
Founding cohort rate: $349 instead of $449. These are the first dates, so the price is lower. In return, I ask for your honest feedback after the day.
Join once, come back free: you can join any later date at no cost, when a seat is open. Email me and I add you.
At 4:45 pm ET your own loop runs on your machine, with a cost gate and a review sheet, and it keeps running after class.
Write an agent instructions file (AGENTS.md or CLAUDE.md) that fits on one screen, and a handover file.
Add one hook that blocks a mistake you already made, and a guard that stops paid API calls.
Set up a schedule, a job queue, and a worker with launchd or cron.
Run one agent task that writes one findings file while you are away.
Add an allowance check that runs before each job.
Know in one sentence what the terms of your plan allow, and when to use an API key with a dollar cap.
Read one night of agent output in 10 minutes.
Use a blind grade in place of the grades the agent gives its own work.
What the agent reads before it acts. You leave with an AGENTS.md or CLAUDE.md under one screen and a handover file, in your repo.
You leave with one hook that blocks a mistake you already made, and a guard against paid API calls.
A 45-minute break.
Schedule, queue, worker, findings. You leave with a launchd or cron job that runs one agent task and writes one file.
You leave with an allowance check before each job and a stop rule. Anthropic says its plan limits "assume ordinary, individual usage", so we read your plan terms together.
You leave with a 10-minute review sheet and a blind grade that replaces the grades the agent gives its own work.
A queue item goes to research and then to a draft, live. You leave your loop on for Monday. We end at 4:45 pm.

I teach from my own code, bills, and mistakes. 6 years at Amazon.
Engineers who use Claude Code or Codex every day and want the agent to work when they are away.
Tech leads who own the agent setup and the token bill for a team.
This workshop does not fit first-time Claude Code users. Take a beginner class first.
The class starts from a setup that already works. You need a paid Claude or ChatGPT plan you already have.
You run commands and commit to your own repo during class.
The schedule block uses launchd on Mac and cron on Linux and WSL. Optional: an API key with a dollar cap.

Live sessions
Learn directly from Saurav Sharma in a real-time, interactive format.
Lifetime access
Go back to course content and recordings whenever you need to.
Community of peers
Stay accountable and share insights with like-minded professionals.
Certificate of completion
Share your new skills with your employer or on LinkedIn.
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Maven for Teams
Reimbursement
Get your company to pay
Everything L&D needs: email template, receipts, and certificate of completion.
Get reimbursedTeam discount
Learn with your teammates
Save 20%+ when 2 or more teammates enroll in the same cohort.
Save 20%+ with a teamPrivate cohort
Run a cohort for your org
A dedicated cohort with a custom schedule and curriculum, tailored to your team.
Book a private cohort$349
USD
10am–5pm EST
·4 cohorts