Twenty years ago, we started MDCalc by physicians and for physicians with a simple belief: physicians deserve tools they can trust.
Over the past two decades, that belief has guided everything we’ve built. We focused on evidence over opinion, transparency over black boxes, and helping clinicians make better decisions — not replacing their judgment.
Today, medicine stands at another inflection point.
Generative AI is one of the most important technologies our profession has ever encountered. Every day, it becomes more capable, more accessible, and more deeply integrated into clinical care. But history teaches us that every transformational technology introduces new risks alongside new possibilities. If medicine teaches us anything, it’s that there is no free lunch; everything carries risks and benefits.
When anesthesia transformed surgery, it expanded what physicians could do, but it still left half of patients dying from infections. But medicine didn’t abandon surgery; it adopted germ theory and developed hand-washing, anti-sepsis, and sterile technique. These became new standards that made surgery safer than ever before — and now looking back, we’d never allow an invasive procedure to be performed without all of these added layers of safety.
We believe AI is at a similar moment: the question isn’t whether physicians should use AI. The question is how.
We are also worried about what over-reliance on generative AI does to the clinical decision-making muscles built in training and sharpened every day of practice. This isn’t speculation — a 2025 study found that highly-skilled endoscopists who regularly used AI for polyp detection saw their unassisted detection rates fall once the AI was removed. The tool didn’t fail. Their unaided skill did.
We believe AI should sharpen physician judgment, not replace it. A profession that quietly outsources its pattern-recognition to a model — and only discovers the cost when the model is unavailable, wrong, or gone — hasn’t been augmented. It’s been hollowed out. This is how AI actually takes a physician’s job: not by replacing us, but by letting us replace ourselves — one convenient shortcut at a time, until the judgment we trained a decade to build has atrophied, and there’s nothing left that a model can’t do better. CrossCover is built to keep the physician in the loop by design, not as an afterthought.
We don’t believe clinicians should have to choose between using new technology, practicing safely, and staying sharp. We believe they deserve all three.
That’s why we created CrossCover.
CrossCover is built around a simple idea: physicians should be free to use the AI models they trust, while benefiting from an independent layer of safeguards designed specifically for clinical care. We believe protecting patient privacy, surfacing hallucinations, reducing hidden bias, preserving physician judgment, and maintaining transparency aren’t optional features — they’re foundational principles for the future of medicine. When discrepancy arises in LLM output compared to traditional scientific sources (e.g. journal articles, textbooks), features such as CrossCheck (currently in beta) allow clinicians to assess and apply their clinical judgment.
We’re also building CrossCover as an open platform. The future of clinical AI won’t be defined by a single model. It will be built through collaboration among organizations that share a commitment to physicians, patients, and evidence-based care. That’s why we’re excited to work alongside leading AI companies, health systems, researchers, and clinicians to help define what trustworthy clinical AI should look like.
Ultimately, CrossCover is about helping physicians practice with greater confidence in an AI-enabled world.
Just as MDCalc helped clinicians navigate the evidence revolution, we hope CrossCover helps the medical community navigate the AI era with the same commitment to rigor, transparency, and trust.
We’re grateful you’re joining us at the beginning.
Joe Habboushe, MD MBA
Graham Walker, MD
Shazia Siddique, MD, MSHP