How Lucija Tomšić drafts 35 Substack notes a week with 8 agents for about $8 a month

The problem

Five notes a day, seven days a week, adds up to 35 notes on top of the newsletter itself. In her words, “The bottleneck was never ideas, and it was never writing one note. It was volume, rhythm, and timing.” Substack’s scheduler takes one note at a time, which she calls “soul-destroying” at that volume.

She is clear that this is not an autonomous agent: “Mine is a workflow: eight specialized prompts, run in order by a script, with me approving in the middle.” She calls them agents because each has a single job. “But it does not pick its own goals. It follows mine.”

The blueprint (as she describes it)

Fix scheduling first. She built a small tool that bulk-schedules from a CSV. One upload creates all 35 drafts and their posting times, and Substack’s own scheduler does the work underneath.
Write the three core files. These are a voice profile, an “about me” with real stories, and a list of writing habits to avoid. She says these “do 80 percent of the work.”
Run on a schedule. The system runs in the cloud every Wednesday at 6pm, the night before she publishes her article. It emails her when the drafts are ready.
Phase 1: five research agents in parallel.
- Past-Performance Analyst: reads last week’s likes, restacks and replies.
- Article Analyst: pulls 10 to 14 angles from the week’s new article.
- Evergreen Analyst: mines everything she has published for timeless ideas.
- Competitor Analyst: studies top creators in her niche for patterns to adapt (“Adapt, never copy”).
- Audience Analyst: reads her engagement heatmaps for posting times.
Phase 2: plan, then write.
- The Synthesizer turns the research into a weekly plan on a fixed mix: 40% from the new article, 30% evergreen, 20% competitor-inspired, 10% pure brand voice.
- The Writer drafts all 35 notes. Its inputs are the brief, posting times, article and evergreen angles, competitor patterns, her voice file and her “about me”. It is “not allowed to invent my life”, so personal stories must come from details she has written down.
- Each note uses one of seven fixed formats: Imperative Command, Bold Claim plus Reality Check, Definitional Framework, The Confession, Quick Tip, Ultra-Short Truth (under 50 words) and Data Truth.
- Hard constraints: no two notes in a row use the same format, no two notes open with the same first line, the first sentence is the hook, specific numbers always, no emojis, no em dashes, no closing call to action, and a set mix of lengths each week.
Human approval. In the dashboard she reads the brief, edits the notes she wants to change and clicks “approve and push to scheduler”. The engine passes the list to her bulk-scheduler as a CSV, and all 35 drafts are created in Substack at their scheduled times.
Phase 3: the Retro Analyst. After the week ends:
- It scores each note against its own job. Story notes are judged on likes and replies, identity notes on restacks and question notes on replies.
- It compares what the Writer drafted with what she published and treats each hand edit as a correction. It writes the patterns into a running voice file, which the Writer reads next week. The file is capped, and when a new correction contradicts an old one, her latest behavior wins.
- It writes hypotheses for next week, stated as bets, and hands the next Synthesizer a short list of changes.

Her suggested build order is the scheduler first, then one writer prompt with a good voice file, then the other agents as needed. “You do not build all eight on day one. I did not either.”

Stated results

“Last Wednesday it drafted a full week of notes before I woke up.”
About $8 a month, which she breaks down as roughly $2 a run, once a week.
No engagement, subscriber growth or time-saved figures are stated. The “about 80 percent” restack figure in the post is an example of a hypothesis the coach writes. It is not a result.

Caveats

Attribution. The post body does not name its author; the page’s author metadata on brisk.vision names Lucija Tomšić (checked 2026-10-16).
First-person. This appears to be the author’s own blog. It is not a vendor story or interview.
Promotion. The post offers the scheduler template (“the scheduler repo”) to readers who recommend the brisk. newsletter. The full engine stays private. She says she may write up her setup if enough readers ask. Nothing is sold, but the post is a referral push.
Human in the loop. Nothing publishes until she has read it, and the newsletter article itself is never touched by the system.
Her own list of failures:
- The coach over-rates its own work.
- Output drifts toward generic when the week’s article is thin.
- Timing is a guess. Substack doesn’t show when readers are online, so the Audience Analyst uses her past engagement as a proxy.
X: Unverified not checked.

Sources

https://brisk.vision/blog/substack-notes-engine/ (primary: first-person write-up of the workflow)

Want agents like these working your distribution? See the Agent Stack Teardown. Get new teardowns weekly: subscribe (double opt-in).