Running 20 newsletters with a small team: what had to be automated

It is 7 a.m. on a Thursday morning. You have two hours to deadline, a part-time contributor on the keyboard, and this is not the only publication going out today. The same steps run again: open the city feed, the school district, the shelter, the home listings, the community calendar. Draft each story in the publication's voice. Find photos. Check every link and price before a reader catches the mistake. That is the moment I remember most from running our own portfolio, and it is the moment that made us admit a browser full of tabs was not going to scale.

We run more than 20 local publications with 175,000+ subscribers. I am not going to pretend the problem was exotic. It was arithmetic. Multiply the same two-hour production run by a growing stack of towns and the week stops working. The question was not whether to automate. It was which steps could honestly be handed over, and which ones had to stay with a person.

The short version

Local newsletter production breaks into repeatable steps and judgment calls. The repeatable steps are sourcing feeds, drafting stories in a consistent voice, making GIFs for homes, pets and restaurants, and fact-checking the finished issue. Those can run in software. The judgment call, deciding that this issue is right and sending it, belongs to a person. The manual workflow usually takes 2 to 3 hours per issue; once the repeatable parts run in software, most of our publications ship in under 15 minutes from source to send.

What got automated first: the sourcing

Every local publication lives or dies on the same question: what happened in town today? Answering it by hand means hunting through a pile of open tabs before writing a word. The fix is to check the sources every morning, automatically, so that when you sit down the stories are already waiting. The sources that matter are consistent from town to town: news from the city, schools, county and police; shelter pets; new restaurants and businesses; home listings; community calendars; local jobs; local video channels; and any website or RSS feed a publisher wants to add.

The same is true of the formats readers recognize. A home listing becomes a Featured Home with the price, beds and baths. A shelter animal becomes a Pet of the Week. A new place becomes a Restaurant Radar entry. Each of those can carry a moving GIF built from the listing's or the place's own photos, made without anyone opening an image editor. Drafting in the publication's voice is the same kind of repeat: the voice settings are fixed, the listing details are already structured, and the writing can be produced to spec.

What stayed human: the send

Fact-checking looked like the least automatable step and turned out to be one of the most. A review pass that reads the whole issue against every source it came from, links, photos, prices and dates, and lists what to fix, removes the slowest and most error-prone part of the morning. On the September 29 Leander Scoop issue, that pass opened 31 links and fact-checked 18 stories in 27 seconds. But it only lists what to fix. Accepting the fixes and pressing send is still a person's call, and we have never found software that should make it for us on a Thursday morning.

The tool we ended up building

We could not find a tool that solved this production problem, so we built one, and it now exists as lightbreak, at app.lightbreak.ai. It does the steps described above: checks a publication's sources every morning, writes each story in the publication's voice, makes GIFs for homes, pets and restaurants, runs the review pass over the whole issue, and pushes the finished draft to Beehiiv or ActiveCampaign, where subscribers, delivery history and list health stay put. Moving over costs nothing, and a person still decides when an issue sends.

If you run tools like Claude or ChatGPT alongside your workflow, they can drive the same pipeline over MCP, a protocol that lets an AI assistant call the app's tools: build an issue, write sections, make GIFs, run the review and push a draft. A key is created in Settings under AI agents and API. The review still runs, and the send still waits for a person.

Who this fits, and who it does not

This fits people who run local newsletters, one publication or a portfolio, often with a small or part-time team working to a deadline. It also fits real estate agents and other local businesses who want a high-quality newsletter with featured homes, local events, local news and restaurants. If your email platform is Beehiiv or ActiveCampaign, nothing moves. If you are on another platform, this tool does not connect to it today, and that is a fact worth checking before anything else.

It does not replace an editor. The voice settings make each story sound like your publication, but the review agent's list of fixes still needs a person to accept them, and the send button still needs a person to press it.

How to start using it

Questions worth asking first