Anthropic just filed for IPO 🚀 Funnily enough, their IPO's page 3 has a picture of a plate. Reshape Colony Counter App: - 0.8s - 74 colonies - pretty much dead on Claude Opus 5.5: - 27 minutes(!!!!) - 66 colonies - Crafted +50 sub-images and still missed a few. I've no doubt their IPO is going to be massive, it's still interesting how poorly Claude performs on something they're posting on the frontpage of their filing 😅 Biology is solved right?

Closing that gap is what we’re building at Reshape: robots that grow and watch living microbes around the clock, and AI that helps decide what to run next, from the first plate to the fermenter.

AI can design a thousand proteins before lunch. Proving one works still takes weeks, months and even years to bring to production.

But almost everything biology makes at scale, from enzymes to medicines to food protein, is made by living cells in a tank. Cells are slow, fussy and expensive to grow, and nobody has drawn this curve for testing them. That’s where most good designs die, somewhere between the lab and the factory.

I can’t believe some of our competitors are still doing product demos with pictures in slides. Tech has progressed so much the last year, the divide between good and great companies is just going to widen. So, to help and inspire others: Here’s the demo of our world-class, 700 plates/hr, ~4x better AI-than-humans at finding pathogens, validated plate reader.

meanwhile, in bio: "nooo you can't just scale up a small scale fermentation it's way too complex"

75% less salmonella hitting the markets. Of course AI is better at finding pathogens in food.

Microbes in 4D. This is parts of the data feeding out detection algorithms to figure out which organism is which, allowing us to annotate billions of microbes automatically.

Biology is the most complex technology that exists, and today, we're capturing less meta-data than we do for marketing leads. We need a step-change in how biology is done if we want the H100's to have any significant effect.

A timelapse turns it into data. Growth rate of each partner. The day the gap opens. How wide it gets. When the pigment kicks in. Same plate, 100× the information. This one is 338 images, one every 10 minutes, from one of our smart incubators.

The catch: most co-culture work ends with one photo on day 7 and a caliper. "Inhibition zone: 4 mm." Everything that happened on the way there is gone.

It's also how you build microbial communities on purpose. Cheese rinds, sourdough, soil consortia, probiotics.Who tolerates whom. Who dominates. Who only shows up when a partner is there. You can't read any of that off a single strain.

that gap is a molecule with a job. This exact setup is how people find biocontrol strains that keep pathogens off crops, protective cultures that keep spoilage molds out of food, and new antibiotics. A lot of biosynthetic genes only switch on when a competitor shows up.