# How Daria Cupareanu built a Claude Code agent that handles her distribution while she sleeps

Daria Cupareanu writes the newsletter, and her agent handles the distribution. A scheduled Claude Code Routine wakes every six hours in Anthropic's cloud and runs whatever task is due. It files each new article into an internal wiki, drafts Reddit posts and replies, and publishes approved LinkedIn posts. Every draft goes to her Slack for approval, rejection or a rewrite. Every Monday the agent reads her edits and rewrites its own instructions and skills.

| | |
|---|---|
| **Who** | Daria Cupareanu, writer of the newsletter at [aiblewmymind.substack.com](https://aiblewmymind.substack.com) ([LinkedIn](https://www.linkedin.com/in/daria-cupareanu/)) |
| **Problem** | Distribution across Reddit, LinkedIn, Substack notes, Instagram, Facebook, Threads and X. In her words, "None of it is hard, all of it takes time, and it has to happen every single day." |
| **Stack** | Claude Code + Claude Code Routines (scheduled, cloud-run), Gmail, Slack, Claude skills, an internal wiki built on Andrej Karpathy's LLM wiki method, a GitHub repo of plain text files |
| **Confidence** | Medium (first-person post, but the full breakdown sits behind a paywall) |
| **Last verified** | 2026-10-16 |

## The problem
Daria already had a Cowork OS for the work she wants to stay "deep inside": writing the newsletter, client projects and Amplifiers. Distribution was the work she dreaded. It meant turning each article into Reddit posts for each subreddit and answering threads where people ask what she just covered in >2,000 words. It also meant drafting LinkedIn versions with infographics, pulling lines into Substack notes and repurposing for other platforms. She knew Reddit mattered for growing the newsletter and for getting indexed by LLMs, and that consistent LinkedIn posting mattered too. "I just dreaded both, so the work stayed undone."

## The blueprint (as she describes it)
1. **Draw the line first.** The newsletter writing stays hers. "Distribution gets automated, but nothing publishes without me."
2. **Run it on a schedule in the cloud.** A Claude Code Routine wakes the agent every six hours and it runs whatever task is due. She chose Claude Code over Cowork for this reason: Cowork's scheduled tasks "run on my computer, with the app open," while Routines run in Anthropic's cloud.
3. **Ingest each article.** When she publishes, the agent reads the new article from Gmail, saves it and files it into her internal wiki, which is built on Andrej Karpathy's LLM wiki method.
4. **Repurpose for Reddit.** It drafts a Reddit post from the article and suggests which subreddits fit.
5. **Answer Reddit threads.** It scans subreddits for questions her work already answers, which it knows from the wiki, and drafts replies.
6. **Route every draft to Slack.** She approves, rejects or rewrites each one. An approval tells the agent it is on track. A rewrite shows it what she would say and how. A rejection tells it the thread was not worth answering.
7. **Post approved LinkedIn posts.** It publishes approved posts, text and image, on schedule and tells her when they are live.
8. **Tune weekly.** Every Monday it reads all her edits from the week and rewrites its own instructions and skills to sound and think more like her.

She adds one workflow at a time. She does a task by hand until she learns how Claude handles it, then hands it over and reviews the drafts. The whole system is "just a folder of plain text files," kept separate from her Cowork OS.

## Stated results
- The system has been running for over a month.
- She says the weekly hours saved are "the smaller win." No hours figure is given.
- Reddit and LinkedIn work she had avoided for months now happens daily, and "the part that needs me takes minutes."
- No traffic, subscriber, engagement or Reddit karma figures are stated.

## Why this matters for postby.ai
- Her rule is a useful filter for choosing what to automate: "start with the tasks you dread." Her reasoning: "Automating work you’d do anyway saves you hours. Automating work you keep avoiding gets it done at all."
- Approve/reject/rewrite in Slack doubles as training data, which the weekly tuning run then uses.
- Cloud scheduling decides whether a system can run while the laptop is closed. She says it is the technical reason she used Claude Code instead of Cowork.

## Caveats
- The post sells a paid tier. The full breakdown of the six automations, the folder structure and memory files, the wiki, the self-learning loop and her watch-out checklist are for premium readers only. This teardown covers only the free section.
- She also mentions plans to package articles into paid products on Gumroad.
- She refers to a GitHub repo, but the link is not in the source text.
- Not everything is automated. She polishes every draft "until it sounds like me before it goes out." She still does 20% of the Instagram, Facebook and Threads repurposing by hand and is not yet happy with the carousel copy.
- She says "nothing publishes without me" and also that "the posts that are safe to publish keep going out on their own." The LinkedIn posts that go out are ones she has approved.
- Planned work (research layer, X, lead magnets via ManyChat, products) is not built yet according to the post.

## Sources
- https://aiblewmymind.substack.com/p/claude-code-content-distribution-ai-agent (primary: Daria's first-person post, Part 1)

---
*Want agents like these working your distribution? [See the Agent Stack Teardown](/agent-stack-teardown). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*