When standard medical care fails, most people accept their fate. When you are the co-founder of a multi-billion dollar open-source company, you treat your body like a legacy codebase—and you start debugging.
In a recent, deeply moving OpenAI Forum event, Sid Sijbrandij (Co-founder and Executive Chair of GitLab) and Jacob Stern (Geneticist) sat down to explain how they leveraged advanced diagnostics, biological engineering, and ChatGPT to fight a rare and aggressive form of bone cancer: Osteosarcoma.
This isn't just an inspiring story. It is a masterclass in treating biology as an engineering problem. Here is the technical breakdown of how they did it.
🛑 The Walled Garden of Modern Medicine
Sid’s journey began with a misdiagnosed pain during a bench press. It escalated into emergency spinal surgery to remove a six-centimeter tumor. He endured radiation, severe chemotherapy, and four blood transfusions just to stay alive.
Despite standard treatments and even experimental click-chemistry therapies, the cancer came back. His oncologist had no more standard-of-care options.
The traditional medical system operates on a very specific incentive structure: Doctors want to minimize liability. They prescribe treatments vetted by $100 million randomized controlled trials. But for a rare cancer patient, that timeline is a death sentence.
"In my case, because it's a nasty cancer with medieval medicine, I'd rather die from a treatment than from the cancer... That is not what you will get with doctors. They have very different incentives than you." — Sid Sijbrandij
Sid needed to switch his objective function from minimizing liability to maximizing survivability.
💻 Going into "Founder Mode"
Sid Sijbrandij discussing his decision to go "Founder Mode" on his cancer.
With no options left, Sid quit his day job and went into "Founder Mode" against his cancer.
Instead of waiting for a doctor to guess the right treatment, they initiated Maximal Diagnostics. They ran every test under the sun—single-cell sequencing, bulk RNA sequencing, targeted radiodiagnostics, and organoid models—generating over 25 Terabytes of data on Sid's specific tumor.
But raw data is useless without a parser.
🧠 The "Iron Man Suit": ChatGPT in Practice
Analyzing 25TB of data requires massive compute. Jacob Stern used ChatGPT to democratize that analysis.
This is where Jacob Stern, a geneticist formerly at 10x Genomics, stepped in.
Analyzing 25TB of biological data typically requires a small army of specialized bioinformaticians. Instead, Jacob used ChatGPT as an "Iron Man suit" to democratize that specialized knowledge.
The RNA Sequencing Breakthrough:
Jacob fed the raw output of a bulk RNA sequencing experiment (a massive CSV file mapping genes to counts) into GPT-4, simply asking: "What do you think of it?"
The AI instantly flagged B7H3, an abnormal protein expression, and mapped out complex immune dynamics. This wasn't a final diagnosis, but it was a rapid, highly accurate hypothesis generator.
Building an agentic swarm to analyze blood cells dynamically.
Agentic Bioinformatics:
They didn't stop at simple chat prompts. They built an automated harnessing system—a swarm of AI agents.
When they suspected a dangerous chemotherapy side-effect called CHIP (Clonal Hematopoiesis of Indeterminate Potential), Jacob didn't wait weeks for an expert consult. He asked their AI system in natural language. For about $20 in API costs, the AI:
- Conducted a 30-minute literature review.
- Formulated a hypothesis and selected biological markers.
- Wrote and executed Python code directly against 600,000 single cells sequenced from Sid's blood.
- Returned interactive plots and a conclusive report.
The AI didn't replace the doctors. It accelerated the engineering team so they could speak to the specialists at an expert level.
⚙️ The Technical Fix: An "AND Gate" for T-Cells
Programming a logic gate into CAR-T cells to spare healthy organs.
With the data parsed, they needed to build a custom weapon. They designed a CAR-T Cell Therapy—a "super killy nuclear bomb" designed to hunt cells expressing the B7H3 protein that the AI had flagged earlier.
But there was a massive bug in the plan.
A specialized scan in Beijing revealed that Sid's healthy liver expressed 3.5x more B7H3 than a normal person. If they deployed the CAR-T cells, the treatment would destroy his liver before it killed the cancer.
The Engineering Solution:
They went back to the lab and literally programmed a Logic Gate into the T-cells.
From their earlier radioligand therapy, they knew Sid's cancer also heavily expressed a fibrous tissue marker called FAP. His liver, crucially, had very low FAP expression.
They engineered the CAR-T cells to require an AND condition to activate:
IF (B7H3 == TRUE) && (FAP == TRUE) { INITIATE_CELL_DEATH }
By turning a biological treatment into a conditional software function, they isolated the target and saved the healthy organs.
🚀 The Future of Open-Source Medicine
Today, there is no evidence of disease in Sid's body.
He didn't just survive; he documented the entire framework at osteosarc.com and is actively investing in companies (like Thalus and Arden) to put this "N=1" personalized medicine pipeline on rails for future patients.
We often debate the hypothetical limits of Artificial General Intelligence. But right now, today, AI is fundamentally changing the speed at which humans can parse biology, engineer bespoke molecules, and debug their own mortality.
Medicine is no longer just biology. It is a data engineering problem.
Source Attribution: This article is based on the OpenAI Forum event "From Terminal to Turnaround: How GitLab’s Co-Founder Leveraged ChatGPT in His Cancer Fight". All quotes and technical details are captured directly from the original presentation.
Top comments (0)