My free cohort was six days out and nothing was scheduled

Shipped

An AI agent built and scheduled six emails for my next cohort in one sitting.

I run a free five day cohort for people learning Claude Code, the AI assistant I build everything with. It is one email a day, Monday to Friday, each with one job that takes about half an hour, plus a final email on Saturday. The next group starts Monday, October 12.

On Sunday, October 4, I found out none of it was set up. The September emails had gone out fine. October had nothing. Six days out.

So I handed it to Claude, connected to my email platform, MailerLite. In one sitting it:

• Setup the October email drip.
• Pointed every copy to the October signup list.
• Scheduled each one for 9am Eastern on its day, October 12 through 17.
• Checked every one came back ready to send, with no warnings, to the right list.

A very repeatable process that I can automate to happen at a regular cadence.

Broke

Nothing rolls forward on its own, and I assumed it did.

The September run worked, so in my head the cohort was a thing that ran. It was not. It was six emails tied to one list and one set of dates. When September ended, so did it.

Two things the agent could not do for me:

• Turn on the welcome email. MailerLite’s connection lets an AI write and schedule campaigns, but switching on an automation still takes a click in the dashboard. That part took a person.
• Notice in the first place. Nothing was watching for “the next cohort has no emails.” I found it by looking.

Everyone already on the October list was added to the start of the welcome email when it was switched on, so nobody who joined early gets skipped.

Steal this

Anything that repeats on a calendar needs a check for its next one, not its last one.

Where this hides:

• A monthly email that only exists because you set up this month’s.
• A recurring class, call or launch where each round is built by hand.
• A reminder that fired last time and was never set again.

The last one going fine tells you nothing about the next one existing. Ask the AI to look forward, not back: “show me everything scheduled for the next 30 days,” then compare it against what you expect to happen.

If it repeats, check that the next one exists. A clean last run is not proof of a next one.

Worth knowing

Anthropic released Claude Sonnet 5.5 on September 28. Sonnet is the middle of Claude’s three model sizes: cheaper and faster than the top one, Opus, and still strong enough for most real work.

• Same price as Sonnet 5: $2 per million tokens in and $10 per million out. A token is a chunk of text about three quarters of a word long, and it is what AI companies bill by.
• Anthropic says it is 30% faster and costs up to 30% less for most work, mostly because it uses fewer tokens to finish the same job.
• Available in the Claude apps on web, iPhone and Android, and to developers through the API.

What changes Monday: not every job needs the biggest model. My nightly video job runs on Sonnet instead of Opus on purpose, because it works through a long checklist rather than thinking through anything new, and Opus was costing about five times as much per run to do the same thing. If you have AI work that is mostly following steps, try it on Sonnet 5.5 first and only move up if the results slip.

Start Monday

The next five-day cohort starts Monday, October 12, and it is free.

It is for anyone who downloaded a Claude Code course and never opened it, or stalled around module two. You do not write code. One email a day, about thirty minutes, and by Friday you have one boring job you used to redo every month running by itself.

Save your seat for Monday. Everyone works through it on the same days, so sign up before it starts.