Why We Built Shadow AI: A Scientist's Story

I spent most of my twenties and thirties in biopharma labs - first at Novartis, then at Momenta Pharmaceuticals. I loved the work. I still do. But somewhere around year six, I started keeping a quiet list of all the things in my week that had nothing to do with science.

The list got long. Searching for old protocols buried in shared drives. Rewriting methods sections that already existed. Formatting conditions tables. Hunting down a reagent lot someone had ordered two years ago. Reconstructing what a colleague had done before they left.

None of that is science. All of it takes time away from science.

The moment the idea landed

There's a specific memory I keep coming back to. I was preparing a study on a Tuesday, and I needed to pull a protocol from a project that had ended eighteen months earlier. I knew the protocol existed. I knew roughly who had written it. I spent most of the afternoon finding it, and when I did, half of it was out of date.

That evening I wrote down a simple thought: the knowledge is here. The problem is the interface. A scientist should be able to ask a question in plain language and get an answer, with sources, from her own organization's work.

From a scientist's frustration to a product

Shadow AI started from that thought. I teamed up with Piyush, who had spent over a decade building large-scale systems at companies like Aetna, Verizon, and Yahoo. He understood how to make AI actually work in production environments, not just in demos.

We set ourselves one rule from day one: the scientist stays in charge. The product's job is to accelerate, not to replace. The scientist defines the problem, makes the decisions, drives the science. The AI handles the searching, drafting, organizing, and preparing.

We call it the 10-80-10 principle. The scientist spends 10% of the time defining and 10% finalizing. Shadow AI handles the 80% in the middle.

What I hope for

I hope the scientist three years from now spends more of her week thinking, designing, and experimenting - and less of it hunting for a protocol from 2022. I hope her team moves faster, learns faster, and gets better medicines to patients faster.

That's why we built this. If any of it resonates with your own week, I'd love to hear from you.