AI workflows
Can AI write and run a local newsletter?
Short answer
Mostly, yes. I automated a local newsletter to about 97% quality in roughly an hour using Claude's multi-agent workflows. The last few percent, and the judgment about what your town cares about, is still a human job.
The day I tested Claude's dynamic multi-agent workflows was the day local newsletter production changed.
The old way was one script trying to do the whole job. The new way is a coordinator agent running sub-agents at the same time: one researches, one writes, one assembles.
What it handled
In one demo, Claude Code pulled adoptable pets from the PetFinder API, found local events, generated images, and wrote the copy for news, real estate, pets, and events, all inside the newsletter platform with nobody drafting. Images for news stories were the one gap.
Events need an agent, not a scraper
Event sites change their formatting constantly, and Facebook and Instagram block scraping. An agent adapts. My target: pull 20 to 25 events to land the roughly eight a newsletter actually needs.
Sources make or break it
Adding sources is the most important setup step, because it decides what news the AI sees. Add the exact page that changes, not just the domain. And check that a source is actually pulling in content. "Added" doesn't always mean working.
The last 10%
AI gets a build 90% done in an hour that used to take months. The last 10% is still hard. That's why I think newsletter production as a service, $500 to $1,000 a month, is real now: the machine does the volume and a person owns the quality.
This is the work lightbreak does for local newsletter operators.
Answered by TJ Larkin. From product and partner calls in May and June 2026. Last updated .