Sid

It's Halloween. Here's a trick-or-treat "thought experiment" for Anthropic (or OpenAI) going big into the Life Sciences. Probably more trick, than treat, but here goes... Both have made making medicines a top priority, as described in Anthropic's S1.  Right now they are taking some “small” bets in the space (~0.1% of expected market cap's).  These are welcome moves and we are hopeful for more engagement in #biotech in general from the frontier lab companies. But if you are a $2T public company (as they are likely to be post-IPO), how could you achieve scale in this life science domain?  What’s the best way for maximum impact on bringing drugs to patients? [assuming that's the goal for their efforts in the life science] I’d argue it's buying a large Pharma company and then radically changing the operating model to put compute and integration in the center. Sort of like retrofitting a big Pharma R&D enterprise into a @NimbusTx model on steroids with compute and tech integration in the center, intercolated with deep biology and drug hunting capabilities, with operational execution on the periphery (impt, but not the driver). Own the learning loops, compress the DMTA cycle across a huge portfolio. Nucleate and empower teams around great potential medicines in a decentralized R&D model... Let me get vaguely specific, for illustrative purposes. Anthropic could buy $BMY for less than 10% of its market cap, an all-stock deal. BMS has about 30,000 people.  Trim commercial (a distraction to the radical change mission) by selling the long list of tail products (top 5 products are 70% of revenues). Then apply frontier AI models to the business across the board - commercial, manufacturing, and R&D. Tons of data sits in these silos. With data, figure out where the bloat is in the corporate bureaucracy and remove it. Be especially aggressive on scaling back G&A and the lawyers, replacing headcount in those functions with AI-enablement.  In the business functions, allow change-makers to focus on operational efficiency. In R&D, BMS has about 3000 scientists/clinicians and operators on their R&D team.  Keep or even grow this group - its where the magic has to start. BMS has a legacy of great R&D advances and has tons of historical data.  Frankly, they probably have some of the deepest datasets on mAb discovery (Mederex mouse one of the most prolific mAb platforms ever), induced proximity (CelMod and related datasets are without peer), cell therapy (Juno was a pioneer with one of largest CART datasets), radiopharma (more recently), and so much more. Deeply mine these datasets. Take all the "me-too" R&D spend on crowded targets and dial it out of the portfolio...  Refocus all that energy on building deep biology datasets to complement the historical modality data. Invest in deep cell and tissue profiling, all the ‘omics, at scale and in an integrated way.  Put AI into the center of the operational workflow as an enabling tool for drug hunters, not replacing them. Work on fewer projects as a whole, but focus on “holy grail” drug targets and first-in-class/first-in-modality programs aiming for step-changes in treatment. And let AI operational efficiency compress clinical timelines as much as possible (10-20%). Unleash teams of drug hunters with AI-powered R&D at scale on any/all therapeutic areas, not restricted to where commercial wants you to focus.  A true science-first enterprise at scale.  “A great medicine makes a commercial franchise, a commercial franchise doesn’t make great medicines” rings true to me here. Commercial voices, I'm told, dominate at BMS, even back in R&D. Maybe that's not optimal? This would be a radical overhaul of a large R&D-intensive company.  Give the experiment 5-7 years to alter the vector of R&D output.  Long enough to see discovery-led changes percolate into early development and clinical PoC... This kind of ongoing R&D investment would be tiny fraction of Anthropic’s revenues during that period, offset in part by the profits from the big commercial brands.  A rounding error for them, but if it worked, this would fundamentally alter the industry’s trajectory. Probably can’t be done - changing large legacy groups is hard - but seems like the only near-term way to truly test “AI at scale in drug R&D” is thru something audacious like this, rather than just thru small (and beautiful) case studies of program-focused biotech … With $5T+ of anticipated market cap between the handful of frontier lab companies, this seems like a relatively small bet on a big prize. And, lastly, make @Prof_Oak_ head of the BMS business unit of $ANT. Trick or treat? Boo!!