# 129 Free Episodes, One Person: How Brent Burgess ships each Sleep Anchor track as five releases with AI and a checked Python pipeline

Brent Burgess runs Sleep Anchor, an ambient sleep-music catalog, on his own. AI models generate each track and its artwork, he masters every track himself, and a Python pipeline renders and publishes one track as five releases: a two-hour YouTube video, an episode on each of two podcast feeds, paid eight- and ten-hour versions for subscribers and a streaming single. Every release is a row in one SQLite calendar, and each irreversible step has a check that can fail, most of them added after something broke. As of Oct 10, 2026 the catalog held 60 tracks, 58 long-form videos and 129 free episodes, and both music feeds have released every Monday and Thursday since late July 2026.

**Try this step:** <strong>One record per track, written first.</strong> A JSON file exists before the track is generated and holds the prompts for the track and its artwork and the AI tools used, so every release traces back to its inputs.


| | |
|---|---|
| **Who** | Brent Burgess, a data leader in California (most recently Director of Data and Analytics at Wine.com), who runs Sleep Anchor alone. LinkedIn and YouTube @SleepAnchor linked from his write-up; Instagram @sleepanchormusic (brand account, name match), verified 2026-11-14 |
| **Problem** | One finished track has to ship to five places with different formats, lengths, artwork and release times, and at the start the steps shared no schedule or state |
| **Stack** | AI music, image and video models; Reaper for mastering; Python and FFmpeg; a SQLite calendar with a local Next.js dashboard; the YouTube API and the RSS.com API; an AI browser agent for steps with no API; Claude Code for much of the code |
| **Confidence** | Medium (first-person, from project records; no audience numbers) |
| **Last verified** | 2026-11-14 |

## The problem
Making a track is not the hard part, he writes. Shipping it to five places is. Each step began as a separate script or browser session, and it showed. The first eight-hour videos repeated the track with no blend, and on one the average viewer left at about the point the track first repeated. A large upload looked stalled and was restarted until five copies of the video existed. A video went public the moment it was uploaded because the schedule setting was off. Publishing a podcast episode needed a browser session each time, and one feed went 24 days without a new episode.

## The blueprint (as he describes it)
1. **One record per track, written first.** A JSON file exists before the track is generated and holds the prompts for the track and its artwork and the AI tools used, so every release traces back to its inputs.
2. **One calendar for every release.** Scheduled videos, social posts and podcast episodes are rows in one SQLite database, with a local dashboard whose buttons start the Python jobs and track them in the same database.
3. **AI generates, he signs off.** A music model generates the track and he picks the take. Python builds the Reaper session from a template, and he adjusts and listens. `verify_master.py` fails a master outside its loudness target or too close to clipping. He hears a test clip across the loop join and approves the looped artwork clip before anything renders.
4. **A check before each step that can't be undone.** The render is sampled for black frames, and a truncated podcast file is deleted. Uploads go to YouTube as private with a publish time, the altered or synthetic content label set, and a duplicate check against YouTube itself. Podcast episodes go through the host's API with a release time and its AI-content flag.
5. **Reconcile against the platform.** Nine scheduled jobs run on the same machine. One posts the first comment when a video goes public, two reconcile the calendar against what YouTube and the social scheduler report, and one flags videos still waiting for manual setup. A release counts as done only when the platform confirms it.
6. **An agent for the forms.** Where a platform has no API, an AI browser agent fills in the web form from a prompt the pipeline generates, and the dashboard tracks it until he marks it done.

## Stated results
- An eight-hour video render went from about 4.8 hours to about 25 minutes once the renderer repeated a short loop.
- A podcast episode went from a browser session each to a render and publish in under ten minutes from the dashboard. Both feeds have released every Monday and Thursday since late July 2026.
- Mastering went from about 50 minutes a track in the first session to a median of about 20 minutes on the 12 most recent tracks.
- 60 tracks, 58 long-form videos and 129 free episodes as of Oct 10, 2026, behind 570 commits since March 2, 2026.
- New YouTube videos run just over two hours instead of eight since late July 2026, after his analytics showed people watched a small part of the eight-hour videos, so almost nobody reached the end and the next video never started.

## Caveats
- No view, listener, subscriber or revenue numbers are given, so the write-up shows the operation, not its reach.
- He listens to every master, so he sets the pace: in July 2026 the podcast schedule went from three episodes a week to two.
- Dark, sleep-graded video can trip the black-frame check, so the upload script has a flag to skip it, set by hand.
- Claude Code wrote much of the code to his design, about 55,000 lines of Python and a dashboard of about 78,000 lines of TypeScript.
- It is a portfolio page: he says he is open to director-level data roles.
- Handles: LinkedIn and the Sleep Anchor YouTube channel are linked from his pages. Instagram @sleepanchormusic matches the Sleep Anchor name and the sleepanchormusic.com site but is not linked from them. No X, Bluesky, Threads, Mastodon or TikTok profile verified.

## FAQ
### How does Brent Burgess publish one track to five platforms?
Each release is a row in one SQLite calendar. Python scripts render each version from one loop file and publish through the YouTube and podcast host APIs on schedule, and an AI browser agent fills in forms where a platform has no API.

### What does AI do and what does he do?
AI models generate the music and artwork, a language model drafts titles and descriptions he edits, and Claude Code wrote much of the code. He picks the takes, masters every track, approves the looped clips and decides the design.

### What broke before the checks existed?
Five copies of one video from restarted uploads, a video that went public on upload, eight-hour videos viewers left at the first repeat, and a podcast feed that went 24 days without an episode.

### Does it grow the audience?
The page gives no audience or revenue figures. It reports output and speed: 129 free episodes and releases every Monday and Thursday since late July 2026.

## Sources
- https://brentburgess.com/work/sleep-anchor (primary: Brent Burgess's first-person project page "One track, five releases: Sleep Anchor's AI and Python pipeline", Oct 11, 2026)
- https://brentburgess.com/ (his site: data leader, most recently Director of Data and Analytics at Wine.com)
- https://sleepanchormusic.com/ (Sleep Anchor's site, run by Brent Burgess in California)

---
*Want agents like these working your distribution? [Get It Built](/get-it-built). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*