Serving bold builders of the future. Not an offer to buy/sell securities or investment advisory services. Disclosures: https://xitter.cf/t.co/NqUCVhdnhA
Global
Joined July 2009
- Tweets9.9K
- Following2K
- Followers133K
- Likes8.8K
Pinned Tweet
Today we're announcing over $9B in capital for a new family of funds — our largest raise in 25 years.
AI is rewriting the rules. We're backing the founders building what comes next.
American companies trying to beat each other on cost could become a distribution advantage for Chinese AI.
“The reality is that there are many American companies using some form of a Chinese open source model internally,” says Lightspeed partner @theamberyang.
As those models get cheaper and more efficient, the pressure to adopt them grows.
With Machiz on Lightwork
Lightspeed retweeted
Today, @JudgmentLabs opens for general access.
We believe long running agent systems will generate more exhaust data than the world can currently process.
Hidden within agent traces are rich records of how to make these systems better: what users ask for, when agents stumble, and how agents interact with the external world.
Judgment is the most performant monitoring and improvement layer for long-horizon agents.
We’re excited to see what you build!
.@Machiz explains why he uses different agents for different tasks.
“I’m also using Muse to change my seat on flights because it seems to be able to get through United’s blockers, but Instinct for some reason can’t.”
Lightspeed Partner @theamberyang is more cautious about sharing personal information, but agents already play a bigger role in her work.
“For my professional stack, it’s a bit more involved, right? I deploy a lot of agents sort of in parallel to help me do research on things.”
Amber sees a potential advantage for Claude Duet in reaching people who already use Claude at work. She and Josh discuss whether that familiarity could encourage more personal use.
Tune in to Lightwork for more!
With agents taking on more of the work, hackers could attempt more breaches at lower cost, including against systems that would otherwise be tedious to navigate.
Lightspeed Partner @theamberyang joins @Machiz to discuss what this means for cybersecurity in the age of agents.
More on Lightwork this week:
→ “Neural Networks Are Already Clean”: @nickcammarata (@GoodfireAI) argues that models are more interpretable than previously thought. Amber breaks down the working hypothesis and why it matters for AI safety.
→ The AI Cost Race: DeepSeek and MiniMax released updates focused on efficiency. Josh and Amber discuss how cost could influence US companies’ model choices and what that could mean for reliance on Chinese open models.
→ Always-on Agents: Claude Duet enters research preview. Josh and Amber discuss what they’re comfortable sharing with personal agents and whether the apps people already use give some companies a head start.
Full episode below.
0:00 Welcome to Lightwork
1:03 AI Agents Hack 100+ Websites
4:09 Hackers Sabotage Supermarket Refrigeration Systems
5:16 Using AI to Make Smart TVs Dumb Again
7:01 Goodfire: Are Neural Networks Easier to Read?
13:29 DeepSeek & MiniMax Race on Cost
18:00 The Personal Agent Race: Duet, Muse, Instinct
22:59 Amber Yang’s Request for Startups
Physical engineering has become a major frontier for verifiable reinforcement learning, yet it still lacks the same evals and verifiable rewards that domains like code and math share.
@XploreAI’s Real World Engineering Bench offers the kind of training signal the models need to help close this gap with:
→ 20 tasks in mechanical, electrical and aerospace engineering.
→ Every run graded automatically, using physics-based verifiers.
→ Each task created by engineers with deep domain expertise.
Xplore aims to build the physical problems models can’t solve yet. Congratulations on the launch @Carlos_Bort, Mike and team!
CC: @JamesAlcorn94, @guruchahal
In the AI race, the price tag matters.
On the latest episode of Lightwork, @Machiz and Lightspeed Partner @theamberyang discuss how the cost of Chinese open models could influence American companies’ technology choices.
“What US company can afford to just use US-made tech when these open models are so cheap and available?” Josh asks.
Amber suggests US labs could continue to lead in closed models while “Chinese open source becomes something that American companies are reliant on.”
She connects that possibility to the broader US–China race toward artificial superintelligence. “And I think that’s why the federal government is taking this very seriously.”
Check out the full episode for more!
With agents taking on more of the work, hackers could attempt more breaches at lower cost, including against systems that would otherwise be tedious to navigate.
Lightspeed Partner @theamberyang joins @Machiz to discuss what this means for cybersecurity in the age of agents.
More on Lightwork this week:
→ “Neural Networks Are Already Clean”: @nickcammarata (@GoodfireAI) argues that models are more interpretable than previously thought. Amber breaks down the working hypothesis and why it matters for AI safety.
→ The AI Cost Race: DeepSeek and MiniMax released updates focused on efficiency. Josh and Amber discuss how cost could influence US companies’ model choices and what that could mean for reliance on Chinese open models.
→ Always-on Agents: Claude Duet enters research preview. Josh and Amber discuss what they’re comfortable sharing with personal agents and whether the apps people already use give some companies a head start.
Full episode below.
0:00 Welcome to Lightwork
1:03 AI Agents Hack 100+ Websites
4:09 Hackers Sabotage Supermarket Refrigeration Systems
5:16 Using AI to Make Smart TVs Dumb Again
7:01 Goodfire: Are Neural Networks Easier to Read?
13:29 DeepSeek & MiniMax Race on Cost
18:00 The Personal Agent Race: Duet, Muse, Instinct
22:59 Amber Yang’s Request for Startups
Are neural networks easier to understand than we thought?
On Lightwork, Lightspeed partner @theamberyang and @Machiz discuss @GoodfireAI researcher @nickcammarata’s working hypothesis that models may be more interpretable than previously assumed.
Amber recalls an analogy from her college classes on small image-recognition models.
“Every layer of the neural net was similar to how your brain would process an image,” she explains.
Using an animal image as an example, she adds, “As each layer processed the image more, you would start to get a better picture.”
Seeing what different layers respond to can help researchers trace how a model arrives at its output.
Josh connects that visibility to AI safety. “We all want to be able to see a little bit more into what these models are actually doing behind the scenes.”
Tune in to the latest episode for more!
With agents taking on more of the work, hackers could attempt more breaches at lower cost, including against systems that would otherwise be tedious to navigate.
Lightspeed Partner @theamberyang joins @Machiz to discuss what this means for cybersecurity in the age of agents.
More on Lightwork this week:
→ “Neural Networks Are Already Clean”: @nickcammarata (@GoodfireAI) argues that models are more interpretable than previously thought. Amber breaks down the working hypothesis and why it matters for AI safety.
→ The AI Cost Race: DeepSeek and MiniMax released updates focused on efficiency. Josh and Amber discuss how cost could influence US companies’ model choices and what that could mean for reliance on Chinese open models.
→ Always-on Agents: Claude Duet enters research preview. Josh and Amber discuss what they’re comfortable sharing with personal agents and whether the apps people already use give some companies a head start.
Full episode below.
0:00 Welcome to Lightwork
1:03 AI Agents Hack 100+ Websites
4:09 Hackers Sabotage Supermarket Refrigeration Systems
5:16 Using AI to Make Smart TVs Dumb Again
7:01 Goodfire: Are Neural Networks Easier to Read?
13:29 DeepSeek & MiniMax Race on Cost
18:00 The Personal Agent Race: Duet, Muse, Instinct
22:59 Amber Yang’s Request for Startups
Even a fridge can be a target for hackers.
Hackers gained access to supermarket fridges in France, though AI involvement remains unconfirmed.
On the latest episode of Lightwork, @Machiz also explores how agents could help people get more out of their devices. One example is a developer who used Claude Code to disable unwanted software on his parents’ TV.
Lightspeed partner @theamberyang explains how agents could make complicated device interfaces easier to navigate. “And it can either be used for good or it can be very scary.”
Josh wonders whether a similar approach could help with his own appliances. “If somebody can rewrite the apps on their TV, I’m sure it’s pretty easy to fix something in a washer dryer.”
Check out the full episode for more!
With agents taking on more of the work, hackers could attempt more breaches at lower cost, including against systems that would otherwise be tedious to navigate.
Lightspeed Partner @theamberyang joins @Machiz to discuss what this means for cybersecurity in the age of agents.
More on Lightwork this week:
→ “Neural Networks Are Already Clean”: @nickcammarata (@GoodfireAI) argues that models are more interpretable than previously thought. Amber breaks down the working hypothesis and why it matters for AI safety.
→ The AI Cost Race: DeepSeek and MiniMax released updates focused on efficiency. Josh and Amber discuss how cost could influence US companies’ model choices and what that could mean for reliance on Chinese open models.
→ Always-on Agents: Claude Duet enters research preview. Josh and Amber discuss what they’re comfortable sharing with personal agents and whether the apps people already use give some companies a head start.
Full episode below.
0:00 Welcome to Lightwork
1:03 AI Agents Hack 100+ Websites
4:09 Hackers Sabotage Supermarket Refrigeration Systems
5:16 Using AI to Make Smart TVs Dumb Again
7:01 Goodfire: Are Neural Networks Easier to Read?
13:29 DeepSeek & MiniMax Race on Cost
18:00 The Personal Agent Race: Duet, Muse, Instinct
22:59 Amber Yang’s Request for Startups
Kevin Weil wants to bring AI’s coding revolution to drug discovery.
“If we can get more drugs to more people more quickly, we solve more disease. We save more lives.”
His mission @warpsciences is to build the labs and infrastructure to make that acceleration happen in America.
“We need to do it. We need to do it here.”
The Investment Memo with @kevinweil
With agents taking on more of the work, hackers could attempt more breaches at lower cost, including against systems that would otherwise be tedious to navigate.
Lightspeed Partner @theamberyang joins @Machiz to discuss what this means for cybersecurity in the age of agents.
More on Lightwork this week:
→ “Neural Networks Are Already Clean”: @nickcammarata (@GoodfireAI) argues that models are more interpretable than previously thought. Amber breaks down the working hypothesis and why it matters for AI safety.
→ The AI Cost Race: DeepSeek and MiniMax released updates focused on efficiency. Josh and Amber discuss how cost could influence US companies’ model choices and what that could mean for reliance on Chinese open models.
→ Always-on Agents: Claude Duet enters research preview. Josh and Amber discuss what they’re comfortable sharing with personal agents and whether the apps people already use give some companies a head start.
Full episode below.
0:00 Welcome to Lightwork
1:03 AI Agents Hack 100+ Websites
4:09 Hackers Sabotage Supermarket Refrigeration Systems
5:16 Using AI to Make Smart TVs Dumb Again
7:01 Goodfire: Are Neural Networks Easier to Read?
13:29 DeepSeek & MiniMax Race on Cost
18:00 The Personal Agent Race: Duet, Muse, Instinct
22:59 Amber Yang’s Request for Startups
Lightspeed retweeted
Zetwerk is now listed on the NSE and BSE.
To everyone who helped build Zetwerk, thank you. We hope you feel proud today.
To our new shareholders, welcome. We have a lot more to build.
Here’s to building an enduring manufacturing company from India, for India and the world.
What can AI learn by watching a chemist work?
Kevin Weil explains why traditional large language models run into problems in physical sciences:
“It's still an area where frontier AI models are not yet changing the world in the way models are changing the world in software engineering and math.”
@Bloomberg spoke with @warpsciences co-founder and CEO @kevinweil about plans to use wearable cameras and pressure-sensing gloves to capture lab work and generate training data for AI.
Read the full story: bloomberg.com/news/articles/…
Lightspeed retweeted
An exciting day for the @zetwerk team.
Eight years ago, @acharyaamrit and @srinathramki set out to restructure global manufacturing around reliability.
Today, Zetwerk is going public with the same commitment to on-time and in-full delivery. When a customer’s order arrives in full on the date promised, the many years of waiting on suppliers shrinks and once scattered capacity becomes plannable.
From a young marketplace to a manufacturing ecosystem now serving customers in 29 countries. Congratulations on the listing Amrit, Srinath and team!
@LightspeedIndia @lightspeedvp
A huge congratulations to @zetwerk on going public!
From electronics to renewable energy, Zetwerk connects businesses with the right manufacturing partners and helps coordinate production across complex supply chains.
Since we first partnered with co-founders @acharyaamrit and @srinathramki in 2019, Zetwerk has grown into a global manufacturing platform serving customers across 29 countries.
Proud to celebrate a well-deserved milestone for the entire Zetwerk team!
CC: @bsomaia
A pair of earbuds is a few grams of plastic, metal and electronics, made across several factories before it reaches a pocket. Solar panels, circuit boards and wind turbines are built the same way.
For years, buyers had to juggle those factories themselves, while the one with exactly the right machine often never heard about the order. ZETWERK (@zetwerk) set out to close that gap.
We first partnered with Amrit Acharya (@acharyaamrit) and Srinath Ramakkrushnan (@srinathramki) in 2019. Since then, that idea has grown into a network of nearly 7,000 suppliers, serving customers in 29 countries across energy, precision products and capital goods. Along the way, Zetwerk has brought work to builders from workshops in India to factory floors in Germany. Amrit and Srinath were joined by Vishal Chaudhary, Ankit Fatehpuria and Rahul Sharma in their journey.
Today, Zetwerk’s name goes up on the NSE and BSE. Behind that name are thousands of people who built it, one part at a time.
Congratulations Amrit, Srinath and the entire Zetwerk team. This film is for all of you, and everyone who builds with you.
@bsomaia
Lightspeed Management Company, L.L.C. (”LSVP”) and Lightspeed India Partners, L.L.P. (LSIP) are distinct investment advisory entities, are not a unitary enterprise and operate independently of one another. From time to time LSVP may draw on its relationship with LSIP, and vice versa but only to the extent consistent with its status as a separate investment adviser.”
Lightspeed retweeted
(1/6) Nirva just raised $33M to bring jewelry with perspective to the world. 🎉
$25M pre-Series A on top of our $8M seed, backed by @lightspeedvp, @southpkcommons and more.
CNET Best of CES 2026. 9,000 on the waitlist. Orders open this winter. First pieces ship this year.
Kevin Weil on why he left OpenAI to build autonomous labs: the next leap in AI needs experimental data the models don’t have.
“The models are data poor in the physical sciences.”
In mathematics, you can push forward inside a computer. In chemistry and biology, you need experiments.
“The way to solve this problem is to generate data at scale.”
“Basically what we’ve done in manufacturing we’re beginning to do as a country with biotech and pharma, except it’s not too late.”
“We can bring it back, but what will it take to bring it back? We need automation and robotics.”
@kevinweil on Lightspeed’s The Investment Memo
While AI has advanced in mathematics, we believe there’s still a major gap in the physical sciences.
@warpsciences co-founders @kevinweil and Solomon Reisberg joined Lightspeed Partner @guruchahal to discuss how their labs could generate the experimental data needed to advance scientific AI.
The company recently emerged from stealth and is building US chemistry labs to make research compounds for biotech and pharma companies.
“First be the hands, then build the nervous system, and then over time become the brain,” says Warp’s CEO Kevin.
The plan starts with expert chemists making molecules for customers. Cameras and sensors would record the work, helping Warp automate tasks and train models to improve how compounds are made.
More capable labs could let researchers pursue ideas for new medicines that time and cost currently put out of reach. The experiments could also produce more evidence for scientific AI to learn from.
Thank you to Kevin and Solomon for joining us!
Full episode below.
0:00 The Future Of Automated Labs
0:43 Introducing Warp Sciences And Founders
1:48 Kevin's Journey And Sol's Background
5:03 Building Warp's Founding Team
10:54 What Is A CRO And How Drug Discovery Works
13:24 The Pain Points: IP, Single Supplier And Geopolitical Risk
15:58 How Warp Is Taking On The Challenge
21:45 Threading The Needle On IP And Data Sharing
25:14 The Future Of Automated Labs, Business Plan And Funding
33:21 Closing Thoughts
“You can scale a company past $100M without having to add more people to your finance team”
That’s what @santinestea is solving for with @dualentry.
“Think of an accounting software as the ultimate report card for a business… what DualEntry enables is that you can get that data in real time”
Santiago said that after their recent Series A raise, “we started opening new verticals and started hiring new AE’s to go and spread the word.”
Lightspeed retweeted
AI can dream up thousands of new molecules. but who is actually going to make them?
that’s a major question for the future of drug discovery, and exactly what @warpsciences is tackling.
the team is building an automated chemistry lab in the US to make that process faster and smarter, helping biotech companies turn AI’s ideas into the molecules they need. it’s quite the business model.
@kevinweil and Sol (two of Warp’s co-founders) joined @guruchahal on the investment memo for a fascinating conversation about robots, physical sciences, AI, and how Warp aims to intertwine them.
the strides being made in AI and science continue to amaze me.
While AI has advanced in mathematics, we believe there’s still a major gap in the physical sciences.
@warpsciences co-founders @kevinweil and Solomon Reisberg joined Lightspeed Partner @guruchahal to discuss how their labs could generate the experimental data needed to advance scientific AI.
The company recently emerged from stealth and is building US chemistry labs to make research compounds for biotech and pharma companies.
“First be the hands, then build the nervous system, and then over time become the brain,” says Warp’s CEO Kevin.
The plan starts with expert chemists making molecules for customers. Cameras and sensors would record the work, helping Warp automate tasks and train models to improve how compounds are made.
More capable labs could let researchers pursue ideas for new medicines that time and cost currently put out of reach. The experiments could also produce more evidence for scientific AI to learn from.
Thank you to Kevin and Solomon for joining us!
Full episode below.
0:00 The Future Of Automated Labs
0:43 Introducing Warp Sciences And Founders
1:48 Kevin's Journey And Sol's Background
5:03 Building Warp's Founding Team
10:54 What Is A CRO And How Drug Discovery Works
13:24 The Pain Points: IP, Single Supplier And Geopolitical Risk
15:58 How Warp Is Taking On The Challenge
21:45 Threading The Needle On IP And Data Sharing
25:14 The Future Of Automated Labs, Business Plan And Funding
33:21 Closing Thoughts
Lightspeed retweeted
Robotic factories for chemistry? Sign me up 🙂
@kevinweil and i go back to the very start of our careers. today, he and his team launch @warpsciences
It was great to revisit those early days and learn more about Warp as I walked him and Sol through the actual memo behind our investment.
We dove into how better lab infrastructure could give scientists more capacity to pursue ideas they don’t have the resources to test today, and the future of science.
Check out the full episode below!
@lightspeedvp
While AI has advanced in mathematics, we believe there’s still a major gap in the physical sciences.
@warpsciences co-founders @kevinweil and Solomon Reisberg joined Lightspeed Partner @guruchahal to discuss how their labs could generate the experimental data needed to advance scientific AI.
The company recently emerged from stealth and is building US chemistry labs to make research compounds for biotech and pharma companies.
“First be the hands, then build the nervous system, and then over time become the brain,” says Warp’s CEO Kevin.
The plan starts with expert chemists making molecules for customers. Cameras and sensors would record the work, helping Warp automate tasks and train models to improve how compounds are made.
More capable labs could let researchers pursue ideas for new medicines that time and cost currently put out of reach. The experiments could also produce more evidence for scientific AI to learn from.
Thank you to Kevin and Solomon for joining us!
Full episode below.
0:00 The Future Of Automated Labs
0:43 Introducing Warp Sciences And Founders
1:48 Kevin's Journey And Sol's Background
5:03 Building Warp's Founding Team
10:54 What Is A CRO And How Drug Discovery Works
13:24 The Pain Points: IP, Single Supplier And Geopolitical Risk
15:58 How Warp Is Taking On The Challenge
21:45 Threading The Needle On IP And Data Sharing
25:14 The Future Of Automated Labs, Business Plan And Funding
33:21 Closing Thoughts