Why I’m building Knotocol

In the last few years, AI has become remarkably good at the parts of biology that happen inside a computer. Protein structure prediction keeps improving. Molecular docking keeps getting faster and more accurate.

And because good ideas now arrive faster than anyone can test them, robotics and automation have followed, to speed experiments up and to run them at scale.

That is real progress. But it quietly assumes something every bench scientist knows isn’t true: that once you know what to test, running the experiment is the easy part.

In practice, a lot of experiments only begin after a great deal of manual troubleshooting. Someone has to find the conditions that make it work.

It usually goes like this. We buy a commercial kit and get a protocol. We read it, work out the conditions for our own samples, and search online for the conditions other people used. Only then do we begin.

Then the run itself happens, and things come up that no protocol mentions. You adjust, you try again. An experiment rarely works the first time.

And that is before the everyday experiments, like Western blots, PCR and cloning, that still trouble so many people in so many labs.

All the knowledge it takes to get an experiment working stays inside each lab: the conditions that held, the ones that didn’t, what to try next. It lives in notebooks, in habits, in the person at the next bench.

Papers publish the procedure that worked. The protocols that failed are never published.

So every lab learns the same lessons again, on its own.

If experiments are where science gets stuck, then making experiments go better is where I want to start. Notos is an experiment companion, a small assistant that works with you at the bench.

It reads the protocol with you before you begin. During the run it walks you through each step, keeps your timers, and listens when your hands are full. And it records what actually happened: every observation, and every condition you changed along the way.

You still run the experiment. Notos helps you read it, run it, and remember it.

Every run Notos records adds a little to something larger. Over time, the records grow into a database of real experiments: the actions people took, the conditions they used, and the outcomes they got, including the runs that failed.

Every run adds a little to something larger.

That database is what will let everyone’s experiments go more smoothly, and faster. And slowly, it becomes the foundation for something that doesn’t exist yet: a world model of the laboratory, one that understands how experiments actually behave.

With a model of the bench, the loop can close. Large language models are learning to plan experiments. Those plans reach the bench through Notos, where scientists run them with Notos alongside. What really happened comes back as records, the world model learns from them, and the next plan starts from what the last run taught.

Plan, run, record, learn, and plan again. That loop, from the plan to the bench and back, is what Knotocol is built around.

Run that loop enough times and Notos moves toward something bigger than any single lab: a collective scientific intelligence, in which what one lab learns, every lab can use.

Experiments that work sooner and more often mean science that moves faster. And faster science is, in the end, how we improve human well-being.

That is why I’m building Knotocol.

CJ Yu

Founder, Knotocol