Our weekly newsletter used to take upwards of 4 hours.
I took it over from my business partner not because I wanted to take it away from him, but because I wanted to see if the Claude Code systems I’d been building could fix the process.
Same four hours, regardless of who sat in the chair.
The problem wasn’t the person. It was the process.
The line I won’t cross
If you want a tool that auto-generates a newsletter from links and RSS feeds, this isn’t it. Our newsletter is written by a person, for people.
So what does the AI do? Everything else.
If you’ve written a newsletter, you know the hardest part isn’t the writing. It’s the 90 minutes of pulling events from three calendar systems, copying descriptions, hunting down links, double-checking dates, formatting, and proofreading before you even open a blank page.
That’s the work that AI is perfect for.
The draft I throw away
Here’s the part of my process that surprises people: I have Andy write a first draft, and then I rewrite basically everything.
When I’m stuck, I’ll talk it through in voice mode - the idea, why it matters, what I’m not sure about. Andy structures it into a rough draft. I strip it down to fragments and rebuild. If you compared the two side by side, what survives traces back to what I said out loud.
This isn’t AI writing. It’s AI helping me figure out what I want to say before I start saying it.
The library that compounds
One of the smartest things to come out of this workflow wasn’t planned.
More and more of our events are monthly or quarterly. Andy noticed this pattern and built a library of recurring event descriptions.
When a monthly meetup shows up on the calendar for the second time, the proven description from last time gets dropped in automatically.
No rewriting. No digging through old emails to find “what did we say last time?”
New descriptions get saved to the library automatically after their second use.
Every newsletter makes the next one slightly easier.
Fix it once, fix it forever
Newsletters are partially identical - same enough to feel like they should be automated, different enough that a script would break. That’s what makes them a workflow.
The difference with this setup: every time I fix a thing, we recognize the conditions that caused it and make sure that condition is solved for going forward.
Found a typo pattern? Fixed.
Template headers in the wrong order? Fixed and updated so it won’t happen again.
Missing link that I referenced but never included? Andy catches it before I send.
I used to wait until something was really painful before fixing it. Now I fix problems the moment I notice them and make room for new ones.
The numbers
We went from one newsletter a week, struggling, to two a week in maybe a tenth of the total time.
The actual writing takes about 25 minutes. Reviewing what Andy assembled and approving the send - maybe another 5 to 10 minutes.
The quality went up because I’m spending my time on the part that actually matters: the writing.
And here’s the thing I didn’t expect. Because the barrier dropped so far, I actually enjoy writing the newsletter now.
When the hard part isn’t hard anymore, you do more of it.