Talked to multiple founders yesterday who have been building their products over the last 3 years. Started manual code. They feel behind and slower than companies that started this year with all code written by agents. I have never seen such an order of magnitude difference.

The biggest bottleneck in AI for biology is still the wet lab. AI can generate more ideas than our labs can test. Closing that gap is how we turn better models into faster discovery. Today, we’re introducing X-Automate at Xaira: bringing agentic AI and physical robotics into the lab. 🧵

Today we’re announcing a new collaboration with @nvidia to bring AI to industrial engineering: - faster simulation - open benchmarks and supporting datasets for physics AI - evaluations for agentic AI in CAE and EDA More here: mistral.ai/news/mistral-nvid…

Turning a simulation idea into a working application takes a lot of work. Frontier AI agents are making it easier. 👏 Developers are combining models like OpenAI GPT-6 Astra with NVIDIA Omniverse libraries to prepare scenes, connect simulation tools, test physical behavior, and guide improvements. From humanoid simulators and driving test environments to digital twins and a browser-based space station, see how NVIDIA developers are bringing their ideas to life. Read our blog 👉 nvda.ws/3UAsXIb

there is no moat

lane

30min to generate and run this aero sim using rotating bodies in a new solver our team’s cooking up

Portrait mode launched 10 years ago today. I think it resonated with people because it didn’t feel digital. It felt analog. And this was likely because the inspiration behind its technique went all the way back to oil painting. Sometimes pure math isn’t enough; you need to combine it with insights from the liberal arts. If you recall Portrait mode shipped a month after the iPhone did. This was because this insight about oil painting took a while to figure out. Initially, Portrait mode was seen as a giant math problem. You’d do disparity calculations on the two sensors and output a depth map. The problem was this approach feels and looks digital; it feels fake. In areas without texture, it’s impossible to do this calculation. In areas with ultra fine detail, a lower resolution depth map leads to terrible transitions and a full resolution depth map takes forever to compute. One solution would be to reduce the amount of bokeh so the issues are not as obvious. But then you lose the whole point of the feature, which was to make it look like you had a large, wide-open lens. The solution to this conundrum—the tradeoff between artifacts and the amount of blur— was quite elegant, but it took a while to stumble across. There’s a saying in painting that if you get the eyes right, nothing else matters. Inspired by this principle from oil painting, the idea was to make sure we never got the eyes wrong. We brought in a machine learning block to find the eyes, and draw a mask layer that protected them, then paired that with the disparity depth map. Counterintuitively, the solution was not to decrease the amount of blur, it was to increase it. The additional blur that we could add once the eyes were protected allowed us to hide the artifacts that were create by bad depth calculation. We let the ears go soft, let the nose go a bit soft. This is that wide open look that people loved, even if people didn’t realize that the ears going soft was why. I was thinking about this recently with regard to AI. We’ll be able to do more amazing things… we just need to study the masters who came before us.

Break every problem down to first principles. The old ways no longer serve what the future needs. @MachinaLabs_ founder @EdwardMehr and @ErikPalitsch break down what @SpaceX taught them and how those lessons apply to the hard problems we’re solving today.

JUST IN: Tractian raised $130M at a $1.4B valuation to build sensors that warn factories before their machines break down. Its sensors attach to industrial equipment, tracking vibration, ultrasound, and other signals to detect problems early. -AI identifies faults and recommends fixes -Trained on a massive industrial failure dataset -Used in 2,000+ facilities worldwide -Customers include Coca-Cola, Whirlpool, and Eli Lilly -Also monitors data centers and critical infrastructure Unplanned downtime costs the world’s 500 largest industrial companies an estimated $1.4 trillion annually.

We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence. block.xyz/dessa

Adithya was onsite with a customer developing supersonic aircraft on his second day at Nominal. Now he builds the tools those engineers use every day. Here’s what that job looks like: nominal.io/blog/new-grad-adi…

We launched the first Engineering World Model, SKY-001, this week and it already outperforms all the frontier LLMs in CAD and BIM in speed, cost and precision. Here’s an example of how our model compares to OpenAI’s Astra, Anthropic’s Fable 5.1 and Google’s Gemini 4. These results did not happen by accident. Read our technical blog to find out how we built SKY-001 (🔗 below).

Mantis is an autonomous construction robot developed by London-based All3 to perform tasks on real building sites. It uses AI to place, fasten, finish, and inspect components without human help. Mantis can carry up to 100 kg, switch tools, and work using real building data.

Last month, we demonstrated high-rate alchemy: turning mercury into real, stable gold, beating CERN’s record by a factor of 10. Last week, we enriched isotopes in the world’s most powerful plasma centrifuge, beating the Soviet record that stood for 40 years. Now, we’re building a fusion reactor in San Francisco to achieve record power density.

The forgotten Volkswagen gooseneck trailer In 1974, a small company in Indiana had a wild idea: why should only pickup trucks get to tow fifth wheel campers? The result was El Chico, a camper trailer that hooked onto a hitch mounted on the roof of a Volkswagen Beetle. Because the trailer pivoted right above the car, the little Beetle could turn completely around underneath it, and the trailer itself was meant to hold everything you needed for a road trip, from a kitchen to a place to sleep. For decades, the film was nearly the only proof it had ever existed, and many people assumed it was just a prototype. But a handful turned up years later, and at least one has been fully restored, still hitched to a Beetle. As the inventor’s own son put it, “El Chico is a unicorn of sorts.”

Test engineers are critical.

Multiplayer AI needs to move past the group chat phase. Chats are chaotic, everyone sees everything (creating social & management issues), and they're bad for long projects. Worse, they keep AI trapped inside existing teams, when its real promise is working & helping across them.