She left OpenAI to fix what "normal" blood work misses

Fidji Simo ran Instacart, then led OpenAI's consumer apps. She has POTS, and her own labs came back normal for years. Her new company is betting AI can find what a standard panel can't.

The gist
Fidji Simo left OpenAI in July 2026 to build ChronicleBio, an AI startup studying chronic disease through blood biomarkers.
The company has extracted 153 terabytes of data from 3,500 blood vials, more than triple what GPT-3.5 trained on.
The target disease is POTS (postural orthostatic tachycardia syndrome), which Simo has herself.
The goal is splitting POTS patients into biomarker-defined subgroups instead of one diagnosis that covers everyone.

She ran Instacart, then OpenAI's consumer apps, then quit to work on her own diagnosis. Simo left OpenAI in July 2026. Fortune's exclusive interview traces the move to POTS, a condition that causes her fatigue, fainting, nausea, and headaches. She co-founded ChronicleBio to study it.

POTS patients get told their labs are normal a lot. Standard panels don't flag POTS. Diagnosis usually comes from a tilt-table test and a symptom checklist, and treatment options stay generic because doctors are working from one label covering a wide range of underlying causes.

153 terabytes came out of 3,500 vials of blood. That's more than three times the 45 terabytes OpenAI used to train GPT-3.5. ChronicleBio is feeding that data into AI models built to find patterns a standard metabolic panel was never designed to catch.

The bet is splitting one diagnosis into several. ChronicleBio groups POTS patients by what their blood shows, not by a shared checklist of symptoms, so a drug trial stops averaging results across patients whose only common trait is the same diagnostic label.

A year in, the company has data, not a cure. ChronicleBio has raised $15 million and has no clinical outcome to report yet. Simo's own diagnosis is the test case: whether biomarker grouping turns a "your labs are normal" conversation into an answer.

Why this matters for you

Told your labs are normal but you still feel off? See why a normal blood test doesn't always mean you're healthy and the difference between normal and optimal ranges.

Free · No account · 45 seconds

See what your own panel actually shows

Upload your blood report and see which markers were tested, what they mean, and where normal and optimal diverge.

Analyze My Blood Test →
Any lab PDF or phone photo · Processed in memory, never stored