Triatomic Capital

‘AppliedAI’ in Deep tech, Health tech, & New economy // Helping great entrepreneurs build 'century-defining' technologies & businesses.

[Jeff Huber 🇺🇸 @jhuber]

1/5 When your AI pair programmer starts making sh*t up (or, 'That time that Claude lost his mind')

I built “Code Mower” – an AI coding harness with Claude Code + Codex coding as peer programmers and reviewing each other before anything merges. It was ridiculously productive on my hobby coding project. Then Friday night Claude invented a full-on fake prompt-injection attack and asked me to act on it.

[@dMatrix @dMatrix_AI]

What if your next coding session happened through conversation instead of an IDE?

At d-Matrix, we’re building the infrastructure behind low-latency speech-to-code experiences with heterogeneous AI pipelines and optimized accelerators.

Read what’s next for multimodal AI: https://t.co/AbeJD36uub

[Jeff Huber 🇺🇸 @jhuber]

If foundational AI models want to "figure out biology" – the fastest way to get there is with a ultra-high-throughput reinforcement learning (RL) loop that works in-vivo (aka as close to human biology as possible).

@ManifoldBio is the way – it's the real-life RL loop for biology.

[Jeff Huber 🇺🇸 @jhuber]

Very impressive progress on math, by @axiommathai.

Go @CarinaLHong & Team!

[Jeff Huber 🇺🇸 @jhuber]

For great AI/ML/software engineers – What's your next chapter?

There's life beyond micro-optimizing click-through-rates or cleverly shaving pennies out of a hedge trade.

AI can & will have a massive impact on curing deadly diseases, saving lives, and extending quality of life for the planet. You could (& should) be a part of it. This is a golden age of AI; the future is being created now. Don't waste your precious life and potential on things that won't matter.

There are many noble options, but @ManifoldBio and the opportunity to work with @GlebKuz & @PierceOgdenJ is one of the best.

[Triatomic Capital @TriatomicCap]

We're proud to back @SGRodriques and Edison Scientific (@EdisonSci) in their mission to solve biotech’s biggest bottleneck: scaling scientific talent.

In a new Gradient Descent interview, Sam (CEO/Co-Founder) details how Edison’s specialized AI agents are transforming the biopharma pipeline:

🔬 End-to-End Discovery: Their multi-agent platform, Kosmos, executes full-loop scientific reasoning – from hypothesis to data analysis.

🧬 Validated Impact: The system recently discovered a novel therapeutic approach for macular degeneration, which was successfully validated in the lab and published in Nature.

📈 High-Throughput R&D: Kosmos enables researchers to process data and generate tens of thousands of actionable findings at an unprecedented scale.

While AI accelerates discovery, Sam also highlights the next critical hurdle: modernizing clinical trials to match this new pace of innovation.

Watch:
https://t.co/LlfEwNNbPE
More on Edison Scientific: https://t.co/y1OqoNarSa

[Sam Rodriques @SGRodriques]

One of my favorite podcast appearances, with @l2k. There is a long history of AI overpromising and underdelivering in biotech and pharma -- why is this time different? Super fun episode.

[Axios @axios]

Exclusive: Math AI startup can prove its work https://t.co/tebsJcOuKK

[Carina Hong @CarinaLHong]

Axiom is growing so fast that there isn't a single desk left at our 6000-sqft office

I've seen exciting things in my career. Nothing compares to the energy of this team

The hiring bar is higher and higher. I believe we have the single best hiring pipeline after OAI/Anthropic 🚀

[stash @stash_pomichter]

it’s time…. 🦾🔥

[Gimlet Labs @gimletlabs]

Gimlet has been named to the @Redpoint InfraRed 100, which honors companies building at the cutting edge of infrastructure.

As data centers transform to meet the immense needs of agentic inference, it is an incredibly exciting time to be building in this space.

Check out our Nasdaq photo, and congratulations to all of the companies featured!

[Peter Zhou @xtzhou]

Coworker (@coworkerapp) launched to the public via self-serve today, giving everyone access to enterprise context-aware AI 82% cheaper. Same frontier models with deeper context and efficient routing. We've been using it at @TriatomicCap for the past few months and love it!

[stash @stash_pomichter]

The AgiBot X2 Ultra humanoid, only on Dimensional

Full SLAM, navigation, multi-camera perception, hosted teleoperation, RL training environments, and an agent-native control interface

Developers on DimOS are shipping hundreds of humanoids into production today

Get building 🦾

[Brycent @brycent]

The most expensive part of drug development may not be the experiment at all...

The avg drug costs $2.6B to develop

The majority of that cost comes from repetitive work.

> Same analyses
> Same disconnected datasets,
> Same avoidable mistakes.

That’s why the @EdisonSci <> @Incyte partnership will be significant in the future discovery

It’s an 8-figure partnership, but more importantly, it’s not a pilot.

Kosmos is being deployed across Incyte’s full R&D pipeline, from target identification to translational biology to clinical strategy.

Incyte has decades of proprietary data across oncology, hematology, and inflammation.

Kosmos turns that data into something R&D teams can continuously learn from.

Every conclusion traceable back to a specific paper or line of code.

Which means less starting from Zero, and more knowledge compounding!

[Triatomic Capital @TriatomicCap]

Continued amazing progress by Edison Scientific (@EdisonSci) to build the most powerful AI Scientist -assistant to vastly accelerate therapeutic discovery &amp; development.

Try it yourself: https://t.co/3bvRWDm1y0

[Sam Rodriques @SGRodriques]

We live in a golden age of biology. So why are people still dying from disease?

Because discovery and development move slower than they should.

Today, we’re partnering with Incyte to change that.

Kosmos is now the first agent that can compress months of drug development into weeks, from the earliest stages of scientific discovery through to FDA approval. @Incyte will be the first company to deploy it across their pipeline.

Work that used to take a team of scientists months now happens in weeks.

Patients can't wait, and neither can we.

[Gleb Kuznetsov @glebkuz]

The best place to do it would be somewhere generating diverse data with high throughput, low latency (~week scale), and deep relevance to living biological systems. That place is @ManifoldBio.

We recently opened up 4 new AI roles, and we are adding an RL-focused role soon:
https://t.co/BMlVgCEYFJ

[Matt McIlwain @mattmcilwain]

https://t.co/tSXVbfIV1u

[@dMatrix @dMatrix_AI]

Tomorrow at #IMAPSMemorySummit, Max (Sunghwan) Min from d-Matrix presents:

“Low-Latency Digital In-Memory Compute Packaging for LLM AI Inference: SRAM, DRAM and Beyond”

🗓 May 14 | 4:35 PM
📍 Hyatt Regency, Santa Clara

Come see how DIMC architecture is reshaping AI inference hardware. ⚡️

https://t.co/RY9vcgnBfq

#AIInference #DIMC #dMatrix #Semiconductor

[iPRONICS @iPronics]

Today’s the day!
AI infrastructure is under unprecedented pressure and photonic switching is rapidly becoming a necessity.
Today, Ana González from iPRONICS and Ted Lichoulas from AFL will lead a live webinar on Optical Circuit Switching (OCS).
Join us!
https://t.co/BnYG5rXGwf

[Jeff Huber 🇺🇸 @jhuber]

@krinza The real-life RL loop needed for fastest AI x Bio progress: Manifold Bio (@ManifoldBio; https://t.co/CCMuRpnbUO).

More: https://t.co/8Iymxq1F3R

[Jeff Huber 🇺🇸 @jhuber]

@KobeissiLetter ... and all these fiber optic cables will plug in to @iPronics – "The first rack-ready optical switch built for AI. Reconfigurable in real time to maximize GPU performance, cut energy use, and scale without costly overhauls."
https://t.co/bKxlB6Q8LN