I Fired My Sleep Doctor and Hired an AI

I spent the better part of two years trying to fix my sleep.

Primary care doctor. A prescriber. Probably 10+ appointments between them. Trazodone. SSRIs. A parade of "sleep hygiene" advice I could've Googled. Random devices. Miracle cures. Melatonin gummies shaped like bears.

Nothing stuck. The prescriptions came with side effects worse than the problem. The advice was generic — designed for the average person, not me. And every failed attempt was another 4-6 week experiment where the answer was always the same: "Let's try something else."

Then I tried something different. I gave the problem to an AI.

The Setup

I built a simple data pipeline: Apple Watch → iPhone → database → AI agent. 23 health metrics flowing every day — heart rate variability, resting heart rate, blood oxygen, and most importantly, detailed sleep staging: deep sleep, REM, core sleep, time awake.

I paired this with an Oura Ring for overnight HRV and more granular sleep data. Total hardware cost: stuff I mostly already owned plus a $300 ring.

The AI agent — I call it Kaddie — had one job: analyze my sleep data every single night, find patterns, suggest changes, and track what worked.

No doctor has this. My PCP sees me once a year and gets a snapshot. Kaddie sees every night and never forgets.

What the AI Caught in 48 Hours

Here's where it gets interesting.

I'd started a microdose protocol — a low-dose psilocybin regimen, the kind that's increasingly common in wellness circles. Within two days, my sleep collapsed.

| Night | Total Sleep | What Changed |
|-------|------------|--------------|
| Mar 24 (baseline) | 9.2 hr | No microdose — great sleep |
| Mar 25 | 2.1 hr | Microdose AM + sleep gummy PM |
| Mar 26 | 2.6 hr | Same combo — disaster again |
| Mar 28 (off day) | 8.1 hr | No microdose — immediate recovery |

Kaddie flagged the interaction within two nights. My sleep gummy contained 5-HTP, a serotonin precursor. Psilocybin acts on serotonin receptors. Taking both was flooding my brain with serotonin — racing thoughts, creative ideation that wouldn't shut off, 2-3am bedtimes.

This interaction isn't in any drug database. No doctor would have flagged an over-the-counter sleep gummy conflicting with a microdose protocol. But the data made it obvious, and the AI pattern-matched the mechanism immediately.

The fix took one conversation: skip the gummy on microdose days. Problem solved in 48 hours — not 6 weeks.

The Root Cause Nobody Found

But the bigger insight wasn't the interaction. It was why I couldn't sleep in the first place.

Every doctor treated insomnia as the problem. Kaddie looked at the data differently. My sleep disruption correlated with rumination patterns — a looping mind that wouldn't quiet down. The problem wasn't falling asleep. It was my brain refusing to stop working.

Instead of another prescription, Kaddie recommended inositol — a naturally occurring compound with evidence for reducing rumination. Targeted at the root cause, not the symptom.

That recommendation, combined with magnesium glycinate and glycine, became my locked-in stack. Three supplements. No prescriptions. Under $40/month.

The Results (Imperfect and Real)

Here's a recent week:

| Night | Total Sleep | Deep Sleep | REM |
|-------|------------|-----------|-----|
| Apr 13 | 8.8 hr | 2.0 hr | 2.5 hr |
| Apr 14 | 5.0 hr | 1.0 hr | 1.1 hr |
| Apr 15 | 8.0 hr | 1.4 hr | 2.5 hr |
| Apr 16 | 6.8 hr | 1.4 hr | 1.9 hr |

It's not perfect. April 14th was a rough night — late evening, too much screen time, broke my own rules. The difference is I know why it was bad. The data tells me exactly what went wrong, and the AI helps me course-correct the next night instead of spiraling into another 6-week experiment.

The good nights — 8+ hours with 2 hours of deep sleep and 2.5 hours of REM — are genuinely elite sleep architecture. Without a single pharmaceutical.

The Bigger Idea

This isn't really a story about sleep. It's about a fundamental shift in how we solve health problems.

Traditional medicine runs on statistical significance: "This drug worked for 60% of people in a controlled trial, so try it for 6 weeks." That's valuable for population health. It's terrible for you specifically.

What I stumbled into is 1:1 medicine: continuous data from your body, analyzed by AI that knows your patterns, iterating in days instead of months. It's the difference between a doctor who sees you twice a year and an AI that sees you every night.

Bryan Johnson made this concept famous with his Blueprint protocol — $2M/year, 30+ doctors, every biomarker measured. What surprised me is how much of that approach you can replicate with a $300 ring, an Apple Watch, and an AI agent that costs less than a Netflix subscription.

We're in the earliest innings of this. My sleep experiment is a tiny, imperfect proof of concept. But the pattern it reveals — continuous data plus AI analysis plus rapid iteration — is going to change how we think about personal health.

Not for 60% of people. For you.