Welcome, humans.

First up: we go LIVE in 5 minutes with Alex Lavaee from Microsoft Research’s Catalyst Lab, and he’s bringing a much better demo than “let’s talk about coding agents.” While we chat, an autonomous coding agent will try to build a 3D Subway Surfers-style game in the background.

Alex is teaching test and verification engineering for agentic coding: how you test what an agent builds, verify what it actually did, and make the workflow reliable enough for real engineering. He’ll use Atomic, an open-source verifiable coding-agent runtime he works on, to run the experiment live.

So instead of only talking about the future of coding agents, we’ll watch one work while Alex breaks down the prompts, tests, verification loops, and human judgment that make agentic workflows trustworthy. And yes, we’re very curious whether the game actually works by the end.

Keep scrolling for more about this week’s awesome interview.

THIS EPISODE WAS BROUGHT TO YOU BY…

APEX-Agents: The AI Productivity Index for Agents

See how the latest models rank for jobs like law, consulting, and investment banking.

Mercor's APEX-Agents leaderboard evaluates frontier AI on long-horizon, multistep tasks across economically valuable work.

Built with partners like Harvey, Ramp, and Cognition.

Every model. Ranked by productivity. 

Special Shout to Outshift for sponsoring this episode! 

New on the pod this week…

So Dr. Trevor Martin co-founded Mammoth Biosciences with Nobel laureate Jennifer Doudna and a simple-sounding goal that gets crazier the longer you think about it:

…turn genetic diseases we manage for life into one-time treatments that permanently change the DNA causing them.

Now, Mammoth is using AI to search a universe of roughly 34B proteins, engineer smaller CRISPR systems, and even generate entirely new protein designs.

Trevor thinks sustained progress could get us to a world with no liver or blood genetic disease in the next 5 to 10 years. So yes, we asked him how close “AI cures disease” is to becoming an actual sentence and not a pitch deck.

Here’s our favorite parts:

• (34:49) Could some genetic diseases basically disappear? Trevor says a sustained push could put liver and blood genetic diseases within striking distance of elimination in roughly 5 to 10 years.

• (18:29) AI is moving from searching nature to generating it: Mammoth can feed huge protein-sequence databases into models and ask them to propose entirely new CAS-like proteins, not merely tweak ones we already know.

• (21:37) The data AI medicine is missing: We have mountains of sequencing and protein data. What we do not have is a clean public dataset connecting preclinical experiments to what actually happened in human trials.

• (36:54) Personalized gene medicine is already possible, technically: Trevor says for some genetic diseases, you could build a personalized treatment today if money were no object. The real problem is making that scalable and affordable.

• (49:23) And then we got to lab-grown mini-organs: Organoids can form structures that resemble parts of the brain, which is useful for testing and also an excellent way to make everyone at the table say, “wait, WHAT?”

The part I kept coming back to: AI can already help biology search and design faster, but the hard part is still proving what survives contact with an actual human body.

Why watch this? Because Trevor separates the real bottlenecks from the sci-fi ones. You’ll understand what CRISPR actually changes, where AI helps today, why clinical data is so valuable, and what still has to happen before “personalized medicine” scales beyond a handful of people.

Watch and/or Listen now: YouTube | Spotify | Apple Podcasts

P.S. At 49:23, Trevor casually explains that neuron organoids can grow structures resembling a prefrontal cortex and hippocampus. He then references The Island (IYKYK). Biology is wild, right?

Keep scrolling for a plain-English CRISPR walkthrough, the newest GPT-6 vs. Claude showdown, and three more episodes worth your time.

Additional Resources: CRISPR, translated into human

If the words “guide RNA” and “CAS protein” normally make your eyes glaze over, here’s the version Trevor gave us:

• It started as a bacterial immune system. Bacteria keep short snippets of viral genetic material so they can recognize the same invader later.

• The guide RNA is basically the address. It carries a sequence that tells the CAS protein which matching stretch of DNA to find.

• The CAS protein is the tool. In the simplest gene-editing setup, it cuts the DNA at that location. The cell repairs the cut imperfectly, which can scramble and effectively switch off the targeted gene.

• Delivery is where things get ugly. The liver is comparatively reachable. The brain and muscle are much harder, which is one reason Mammoth searches nature for smaller CRISPR systems that fit into different delivery vehicles.

That is the real trick: the editor can be incredibly precise, but you still have to get it into the right cells. If you want to see what Mammoth is building around that problem, start here.

UPCOMING EVENT: The Neuron IRL in San Francisco

The Neuron is going IRL in San Francisco on November 18 starting at 4:30 PM PT. If you’re in town, come hang out with us for drinks, bites, and a live recording of The Neuron podcast. We’ll have The Neuron community together in one room, plus some special guests and plenty of AI conversation.

Huge thanks to Slack and Alumni Ventures for helping us bring the night to life. It’s free, but space is limited, so RSVP and save your spot.

🎙️ In Case You Missed It…

Four recent interviews and episodes we think you’ll love.

1. The “I’m Fine” problem with voice AI.

TL;DW: Hume AI CEO Andrew Ettinger explains why voice systems can sound human while still missing the meaning hidden in tone, hesitation, frustration, background noise, dialect, and who is speaking. Turning speech into a transcript can flatten the very signals a good voice agent needs.

Why you should watch: If you are building, buying, or simply talking to voice agents, this explains why “sounds natural” and “understands me” are two very different benchmarks.

Watch / Listen: YouTube | Spotify | Apple Podcasts

2. Want to understand what AI infrastructure actually has to do?

TL;DW: CoreWeave’s Chen Goldberg explains why modern AI infrastructure is no longer a pile of GPUs. Compute, networking, storage, cooling, security, and software increasingly have to behave like one enormous computer.

Why you should watch: If agents are going to run longer, use more tools, and handle real work, the systems underneath them matter almost as much as the model.

Watch / Listen: YouTube | Spotify | Apple Podcasts

Have a topic you want to learn about? Request it here!

Subscribe to our YouTube Channel for more!

Subscribe on YouTube to help us bring in more builders, researchers, and guests who can teach you something useful about AI every week.

Stay curious,

The Neuron Team

P.P.S: Love the newsletter, but don’t want these podcast and livestream announcement emails? Don’t unsubscribe. Adjust your preferences to opt out of them here instead.