Teardown No. 89 · Podcasts

11 Views Taught Him Native Video: How Florian Darroman runs podcast ops with 11 OpenClaw agents, and what broke

Florian Darroman runs 11 named AI agents on OpenClaw to handle the work around his Profitable Founder Podcast. OpenClaw is an open-source agent, and he runs it on a Mac Mini. A lead agent called Marc coordinates the others. A nightly pipeline from 12:30 AM to 7 AM schedules X posts with clips, drafts the newsletter, lists guest prospects and writes sponsor DM templates, and he gets a brief on Telegram at 7 AM. He is specific about what broke along the way. Posts with no video got 11 views, agents duplicated work, clips lost their audio, subtitles got names wrong and posts starting with @ were hidden, and each failure became a written rule.

The problem

Darroman hosts the Profitable Founder Podcast, where he interviews bootstrapped founders making $100K to $10M a year. It runs twice a week on YouTube and Spotify. His human team is one video editor. Before the agent setup, his weeks covered guest finding and outreach, prep, recording, show notes, the newsletter, clips, X posts, scheduling, guest research, sponsors, DMs and website updates. He describes this as “15 hours of work around the podcast to support 2 hours of actual recording.” He says the recording is “The only part that actually needs to be me.”

The blueprint (as he describes it)

  1. Put one lead agent over a team. Marc is the lead agent and COO. He runs morning and evening war rooms, coordinates the other agents and sends briefs on Telegram. Bob is dispatcher and ops manager. He runs health checks and either fixes a stuck agent or failed job or escalates it to Marc.
  2. Give each agent a narrow job.
    • Dan schedules 3 posts a day on X with native video clips and replies to posts with episode links. “He never follows anyone. He never DMs anyone.”
    • Claude is the name of his content agent. It reads podcast transcripts, finds the 3 best clip-worthy moments and writes X posts around them. “Every fact gets verified against the transcript.”
    • Adrien extracts clips from YouTube, removes filler words and silence, burns in subtitles and exports the files.
    • Jimmy studies what Starter Story, Greg Isenberg and My First Million do with titles, thumbnails and formats. He finds guest prospects by scanning X and Reddit for founders hitting milestones.
    • Tyler turns episode transcripts into newsletter drafts, which Claude reviews before Darroman sees them.
    • Mona Lisa finds SaaS companies that might sponsor, looks up decision makers and drafts personalized DM templates.
    • Ariane keeps the Notion workspace clean and backs up files.
    • Crawly crawls Reddit, Hacker News, X, YouTube and official docs every night at 12:30AM.
    • Peter builds and fixes internal tools, including the clip extraction scripts and a dashboard.
  3. Run it overnight on a fixed schedule.
    • 12:30 AM: Crawly gathers intel and trending topics.
    • 1:00 AM: Jimmy does competitor research and finds guest prospects.
    • 1:30 AM: Tyler writes newsletter drafts.
    • 2:00 AM: Claude picks the 3 best moments, writes posts and validates them in the Notion content calendar.
    • 2:30 AM: Claude reviews Tyler’s drafts and approves them or sends them back with notes.
    • 3:00 AM: Mona Lisa researches sponsor targets and drafts DM templates.
    • 3:20 AM: Adrien extracts and edits the selected clips.
    • 5:00 AM: Dan schedules the validated posts on Typefully with native video.
    • 5:30 AM: A quality check verifies every handle and confirms every episode link.
    • 6:00 AM: Marc’s war room reviews everything and sets priorities.
    • 7:00 AM: Darroman gets his morning brief on Telegram.
  4. Keep a short human pass. He reads the brief (30 seconds), reviews and sends the newsletter draft (5 minutes), picks guest prospects and sends DMs (10 minutes) and personalizes and sends the sponsor template (5 minutes). Posts go live at 4AM, 4PM and 9PM, and Dan replies with episode links 15 minutes after each one.
  5. Turn every failure into a rule. This is the part he spends the most detail on.
    • No video, no reach. “Posts with no video got 11 views.” He says native video on X gets 100x the reach of text-only posts or YouTube link cards. Now every post has a video clip attached.
    • Duplicate work. Dan scheduled posts that were already scheduled, and Claude reviewed drafts that were already reviewed. The fix was a status system in Notion: Draft, Validated, Scheduled.
    • Silent audio loss. Adrien was processing clips encoded in VP9, and his tools expected h264. Clips looked fine but had no sound, or the subtitles were garbled. They added codec checks, and every clip is verified before processing.
    • Wrong names in subtitles. The transcription model turned “JK Molina” into “GK Molina” or “Jake Molina.” Now Claude writes notes with every proper noun for each clip, and Adrien cross-references them before burning in subtitles.
    • Posts starting with @. X hides these from followers’ feeds and treats them as replies. “We lost thousands of impressions before catching this.” It is now a hard rule in every agent’s instructions.
  6. His advice for starting. Start with one agent on any machine. Install OpenClaw, which he says is open source, free and takes 10 minutes. Give the agent one repeatable task, such as “Read my podcast transcript and write 3 X posts with clip timestamps.” Write a markdown skill file with examples and what not to do. Schedule it with OpenClaw’s cron jobs, log every failure as a rule and add agents once the first is solid.

Stated results

  • “I went from 20 hours of admin per week to maybe 30 minutes.”
  • His morning pass is 30 seconds for the brief plus 5, 10 and 5 minutes for the newsletter, guest DMs and sponsor template.
  • 3 posts scheduled daily with video.
  • Posts with no video got 11 views. Posts starting with @ cost “thousands of impressions.” Neither has an exact before-and-after figure.
  • One recent video “got 30k views in the first 60h.” The post does not attribute this to the agent system.
  • No subscriber, download, sponsor revenue or booked-guest figures are stated.

Caveats

  • All figures are self-reported. The “100x” reach figure for native video is his claim with no source given.
  • The post opens and closes with a pitch for his OpenClaw Lab Community on Skool, with an “early bird discount for the next 100 members.” The price is not stated.
  • The post is dated Feb 15, 2026. OpenClaw and X behavior may have changed since then.
  • The problem section says 15 hours of work around the podcast. The results section says 20 hours of admin per week. The post does not reconcile the two.
  • “Claude” is the name of one of his agents. The post does not say which models power the agents.
  • He still reviews and sends the newsletter, sends guest DMs and personalizes sponsor outreach himself.
  • The post states a goal of 100K followers each on YouTube, X and Instagram “this year.” This is a target, not a result.
  • Handles (verified 2026-11-05): X @floriandarroman; Instagram @florian.darroman; Threads @florian.darroman; TikTok @florian.darroman; YouTube @asyncr0ne; LinkedIn linkedin.com/in/floriandarroman; sites floriandarroman.com and profitablefounder.xyz. Bluesky and Mastodon: none verified. The X icon in the post footer links to a different account than the verified X handle.

FAQ

What software runs Florian Darroman’s agents?

He uses OpenClaw, which he describes as an open-source AI agent that lives on your machine. He runs it on a Mac Mini, with Telegram for briefs, Notion for status and Typefully for scheduling X posts.

What was the biggest lesson from what broke?

He puts posts with no video first: they got 11 views. Every post now carries a native video clip, and he treats that as a non-negotiable rule.

How did he stop agents from duplicating each other’s work?

He built a status system in Notion with three stages: Draft, Validated, Scheduled. He says every agent knows where things stand.

Does the setup increase his audience or revenue?

The post does not say. He reports time saved (20 hours of admin per week down to maybe 30 minutes) but gives no subscriber, download or sponsor revenue figures tied to the system.

Sources


Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.