In partnership with
Hey, Vaibhav here 👋🏻
How many people does it take to get a new drug from idea to working results in mice?
Usually a whole biotech team and a few years.
Cody Tranbarger did it in under 18 months as a one-person company, with Claude and 11 outside labs.
This week we found out Anthropic bought his company.
Plus a Google game that makes itself up as you play, and Alphabet's worst stock day since 2004.
But first:
A word from two tools that do the homework for you.
Presented by The Deep View
Become An AI Expert In Just 5 Minutes
If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.
This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.
Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.
Presented by Gladly
Your Search Bar Is Haunted
Shoppers type exactly what they want and get “0 results found” back. They don’t complain. They just vanish to a competitor.
Gladly understands what shoppers mean, typos and all, and turns dead-end searches into product recommendations.
Back to Claude X Pharma
Claude got a job in pharma
Making a medicine is a long chain: find the target, design the molecule, make it at scale, do the paperwork, run the trials.
Anthropic says Claude is now working along that chain inside real programs at six organisations, including Bristol Myers Squibb, Generate Biomedicines and Enveda.
In lab-dish tests, immune cells made roughly 10 times as many receptors, the proteins they use to spot cancer cells.
At Enveda, manufacturing chemists were spending up to 75% of their time on paperwork.
People with chemistry PhDs, mostly filling forms. (most relatable stat of the week)
Then there’s the part tech twitter is hyping tf out of.
Anthropic bought Mithril Medicine back in July and only disclosed it this week. Tranbarger used Claude to organise contracts, logistics and experiments across 11 outside labs.
His treatment targets LAMA2-related congenital muscular dystrophy, a rare muscle disease in children.
The mouse study was a small pilot, and the team's own report looked at six Claude errors, at least three of which reached outside labs in Claude-drafted emails the team had edited and sent.
Most of this is still preclinical, which means nobody has tested it in humans yet.
Google’s Genie 4 builds worlds that talk back
Google DeepMind announced Genie 4 on Oct 27.
Type "make it rain" mid-scene and it rains. Mumbai in July does that for free, but still.
Google says it also generates sound, so characters talk back.
Only trusted testers have it now, with Google AI Ultra subscribers next and a waitlist for early access.
Google admits details drift the longer a world runs, so the chair you walked past may not be there when you turn back.
My take
Look at what Tranbarger's job was.
Claude took the coordinating and drafting off his plate, and also made mistakes that reached real labs because the checks before hitting send missed them.
At Enveda it went the other way, with Claude catching 7 issues the humans missed.
So what works is a person who knows the field plus an AI, each catching what the other misses.
Meanwhile US students are leaving computer science, down 12.4% this fall.
Coding still helps. Knowing one field well enough to hand AI a whole job, and spot when it slips, helps more.
NEWS Roundup
Google's smartest model comes with a guest list.
Gemini 4 Carbon topped the Artificial Analysis Intelligence Index at 59, but only vetted cyber defenders in Google's Fairwind program can use it. It also makes things up more often on AA's hallucination test: 11%, up from 4% for Gemini 4 Argon.
Mistral gave away a 1-trillion-parameter model.
Mistral Large 4 weights are now free to download under a modified MIT licence, and Mistral says it roughly ties GLM-5.3 as the top open-weight model for agentic coding.
Your ChatGPT plan now comes with pocket money.
OpenAI says eligible Pro and Business subscribers can claim up to $500 a month in API credits to build apps and agents. Unused credits expire.
US students are swiping left on computer science.
The National Student Clearinghouse says CS enrolment fell 12.4% among undergraduates at four-year colleges this fall, in early figures, while health majors grew 6.2%.
Five AI coding tools had hackable built-in browsers.
Forever Security broke the browsers in Cursor, GitHub Copilot, Codex, VS Code and Claude Desktop. Cursor and Copilot have shipped fixes, so update if you use them. Codex and VS Code are still unpatched.
Cool Tools I found this week
Paste your site's URL and it runs 125 free checks on whether AI agents can find, use and pay on your site, and ranks what to fix first.
Describe an app inside ChatGPT, watch it build live, and publish it with logins and hosting included.
A free, open-source app for Mac, Windows and Linux that runs open models like Gemma on your computer, so your PDFs and chats stay on your laptop.
Reply and tell me: if you handed one whole piece of your work to AI, which step would you still check yourself?
Until next time,
Vaibhav 🤝🏻
