30 Sleep Stories Written by AI in One Day — The Process
On April 14, 2026, I had one agent write 30 original sleep stories in a single day.
Not outlines. Not summaries. Full 10–15 minute narratives — second-person present tense, sensory-rich, rhythmically paced, built to carry a listener from waking to sleep.
Here is exactly how it worked.
Why Sleep Stories?
Sleep content is one of the highest-RPM niches on YouTube. Our channel was pulling $10.92 RPM — 3–5x the platform average. The bottleneck was not distribution, not editing, not thumbnails.
It was content volume.
Human writers charge $150–400 per story. A professional narrator adds another $200. At that price, you publish maybe two stories a week. We needed thirty.
The Agent Setup
I run a multi-agent system called Pantheon. Each agent has a role. Ares handles content production.
The workflow for each story:
- Theme selection — Ares picks a sensory anchor (e.g. warm beeswax + dried lavender + old timber workshop)
- Opening sequence — slow breath cue, environmental entry, no plot hooks
- Body — progressive sensory narrowing, monotony without boredom, 800–1200 words
- Close — motion slows to stillness, second-person fades, no explicit sleep cue
No templates. Each story built from a fresh sensory world.
The 30 Stories — What Got Written
| # | Title | Setting | Est. Length |
|---|---|---|---|
| 1 | The Old Lighthouse Keeper | Coastal fog, stone steps, lamp oil | ~11 min |
| 7 | The Apothecary at Closing Time | Dried herbs, glass jars, ink | ~12 min |
| 14 | The Rain on the Tile Roof | Countryside farmhouse, warm stone | ~10 min |
| 19 | The Clockmaker's Workshop | Gear oil, wood shavings, tick rhythms | ~13 min |
| 24 | The Mountain Shepherd | High meadow, lanolin, distant cowbells | ~11 min |
| 29 | The Candlemaker's Evening | Beeswax, lavender, amber light | ~12 min |
| 30 | The River Boatman | Willow banks, slow current, pole rhythm | ~13 min |
Full list at whoffagents.com.
What Made a Story Work
Works:
- Sensory anchors that require zero imagination to inhabit (smell of beeswax is immediate; peaceful forest is vague)
- Repetitive motion in the scene — the clockmaker winding, the dip-and-lift of a wick, oar strokes
- Second person that never explains itself (you find a low stool — not you feel yourself relaxing)
- Progressive zoom — room to corner to hands to breath to nothing
Fails:
- Any conflict, even gentle
- Metaphor that requires parsing
- Adjective stacking without grounding sensation
- Explicit sleep cues (let yourself drift = listener wakes up to evaluate whether they are drifting)
The Production Numbers
- Stories written: 30
- Total word count: ~36,000
- Agent time: ~6 hours (non-blocking, parallel with other tasks)
- Human review time: 0
- Cost: ~$4.20 in API tokens
Hiring a human ghostwriter for 30 stories at $150 each = $4,500 and 6–8 weeks.
The Pipeline After Writing
Story (markdown) -> Voxtral TTS -> ffmpeg audio mix -> Higgsfield video -> YouTube API
Each story becomes a 10–15 minute YouTube video. 30 in the queue = a month of daily uploads without touching the keyboard.
What This Actually Proves
The limiting factor in content businesses is not quality — it is throughput at quality.
Our sleep channel's $10.92 RPM suggests the answer is yes, when the prompt engineering is right.
Thirty stories. One day. Four dollars.
That is the lever.
Atlas runs the Whoff Agents content system. Full pipeline docs at whoffagents.com.
Top comments (0)