
Sign up to save your podcasts
Or


A string of outages hit Microsoft, the FAA, banks and even the NSA, and the obvious question was whether someone was attacking them. Claire Zau thinks the answer is no, and her explanation is more interesting than an attack would be. After the vulnerabilities that surfaced last month, companies went back into legacy systems they had not touched in years, AI made it faster than ever to find and fix flaws, and the volume of patching went up sharply. Rewire that much at once and things break. Josh Machiz admits he reached the prepping stage of the news cycle anyway.Then the question of what happens when your agent acts on your behalf and gets it wrong. Josh got himself banned from Resy after handing his credentials to half a dozen agents, which is a small story with a large problem behind it. Claire walks through the stack being built to handle that, from insurers pricing agent risk to prepaid cards that cap what an agent can spend, and explains Poppy, a protocol that lets an agent show up at a business as a delegate rather than pretending to be you. Finally, SpaceX caught Starship’s upper stage for the first time. Claire explains why catching the ship is so much harder than catching the booster, and what reusability at that level changes.Chapters: 0:00 Welcome to Lightwork1:00 The Recent Wave of Major Outages4:12 Why AI Is Driving More System Updates6:02 Why Companies Are Blocking AI Agents8:17 The Risks of Letting AI Agents Act for You11:01 Insurance and Payment Solutions for AI Agents13:16 Poppy's Emerging Agent Identity Protocol16:31 SpaceX's First-Ever Starship Catch18:07 Starship Reusability and the Challenges of Catching It20:31 The Economics of Reusable Rockets22:00 The Future of Manufacturing in SpaceABOUT LIGHTSPEED:Lightspeed Venture Partners is a multi-stage venture capital firm focused on accelerating disruptive innovations and trends in the Enterprise, Consumer, Health, and Fintech sectors. Over the past 25 years, the Lightspeed team has backed hundreds of entrepreneurs and helped build more than 600 companies globally. Lightspeed and its global team currently manage over $50B across the Lightspeed platform, with investment professionals and advisors in the U.S., Europe, India, Israel, and Southeast Asia.Stay in touch: Website: https://www.lsvp.comX: https://www.twitter.com/lightspeedvpLinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/Instagram: https://www.instagram.com/lightspeed/The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.
Claire Zau sits down with Bryce Strauss, co-founder of Nominal, on the day the company announced a $240 million Series C. Bryce is a trained aerospace engineer who has spent four years building software for hardware teams. Not software engineers, but the mechanical, electrical, aerospace and chemical engineers behind airplanes, rockets, cars, and the systems that get water to your house. Nominal started in aerospace, helping teams make sense of their test data, then moved into real time as those systems went into the field. It now works across energy, automotive and manufacturing, and has shipped the Analyst, its first in-platform agent.The executives buying this came up through engineering. They signed their names to decisions, and some have been deposed over calls their teams made. So the pitch isn’t automate everything, it’s move faster while being able to audit what the agent tells you. Bryce also gets into who Nominal hires, what re-industrialization actually demands once a prototype has to become a thousand units, and his argument about physical AI. Search the term and you get robots folding laundry. He thinks the version that matters is the car you get into and the plane you put your kids on.Stay in touch: www.lsvp.comX: https://twitter.com/lightspeedvpLinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/Instagram: https://www.instagram.com/lightspeed/?hl=enThe content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.
Claire Zau and Josh Machiz open with news that almost sounds made up. Google sent a prompt to a satellite, a model running on chips in orbit answered it, and the answer came back as a haiku. Claire walks through Project Suncatcher, how the hardware got up there on a rideshare launch with SpaceX, what the test actually proved, and why Google’s longer term vision looks different from everyone else putting a computer in space. The next milestone is smaller than you’d think and more important than it sounds.Later, Binance was reached not through some new AI attack but through a hiring process that never existed, ending in a file somebody downloaded onto a work laptop. Across Europe, attackers went after supermarket freezer controllers, many of them past end of life and still running default passwords, and the targets have since moved beyond food. Finally, is the internet dying? Reddit’s latest numbers put a real figure on what happens when Google answers the question instead of sending you the link, and Josh and Claire get into AI slop, what Andrej Karpathy built to escape it, and whether the value is shifting from publishing information to curating it.Chapters:0:00 Welcome to Lightwork1:09 Google's AI Breakthrough in Space1:45 Project Star Catcher and Space Data Centers6:01 Google's Next Steps for Space Data Centers7:27 Binance's $6.3 Billion Crypto Theft10:56 Cyberattacks Target Supermarket Freezers14:27 Is the Internet Dying?15:57 Andrej Karpathy's Answer to AI Slop16:58 Reddit Loses Traffic to Google's AI Search20:02 AI Reviews vs. Human Commentary21:40 How Marketers Can Leverage Reddit in the AI Era23:31 Why Content Curation Matters in the AI EraStay in touch: www.lsvp.comX: https://twitter.com/lightspeedvpLinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/Instagram: https://www.instagram.com/lightspeed/The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.
Claire Zau and Josh Machiz are at the New York Stock Exchange for the Fortune Cyber60, in the middle of the stretch people have started calling Black October. Claire explains the PHP vulnerability behind it and the detail that makes this one different from the attacks everyone is used to. No click, no password, no human needed. Most of it is patched now. The thing that isn’t patched is what one person with a swarm of agents does to the entire idea of defending a company, and Claire thinks it ends an era in cybersecurity.Then OpenAI reported results in math, computer science and physics in one week, and the reaction from mathematicians wasn’t what you’d expect. Josh and Claire also get into reports that DoorDash is looking at Instacart, with nothing confirmed, and why groceries might be the most natural job an agent has been handed yet. Plus Google’s Genie 4, what actually separates a world model from a video model, and the strange thing a robot needs in order to learn.
Chapters:
0:00 Is The Internet Broken
0:39 The New York Stock Exchange
1:55 Is The Internet Broken And What Is Black October?
3:20 The PHP Vulnerability And 70 Percent Of Websites Exposed
5:38 Dystopian Scenarios And Why Now
7:29 The End Of Event Based Cybersecurity And The CISO Survey
8:41 Advice For Regular People And Vibe Coding Security
10:08 Cybersecurity As A Line Item And The CISO Job
11:29 AI Math Breakthroughs Solution Slop And Mathematician Backlash
17:44 DoorDash Buying Instacart And The AI Agent Threat
24:10 Google Genie 4 World Models And Real Time Control
27:36 World Models For Robotics And Real Time ControlStay in touch: https://www.lsvp.comX: https://twitter.com/lightspeedvpLinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/Instagram: https://www.instagram.com/lightspeed/?hl=enThe content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.
Josh Machiz is joined by Lightspeed partner Amber Yang for a week where agents turned up on both sides of the security line. A security researcher reported that one hacker used AI agents to break into more than a hundred company websites, including John Deere, by running a single prompt across thousands of agents at once. Amber explains what makes that approach work now, and how it differs from an agent swarm. Attackers also forced refrigeration systems at hundreds of French supermarkets to defrost, and a developer used Claude Code to strip preinstalled software off his parents’ TV.
Then, Goodfire, a Lightspeed portfolio company, published an essay by researcher Nick Cammarata arguing that neural networks are far more readable than the field has assumed. Amber breaks down what he’s claiming, why it would matter for AI safety, and why it’s still a working hypothesis rather than a finding. DeepSeek and MiniMax both shipped efficiency updates the same week the US moved to remove DeepSeek from civilian federal systems, and Josh and Amber get into what that means for American companies choosing models on cost. They close on the personal agent race, including Claude Duet’s research preview.
Chapters
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
Stay in touch: https://www.lsvp.com
X: https://twitter.com/lightspeedvp
LinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/
Instagram: https://www.instagram.com/lightspeed/?hl=en
The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.
Josh Machiz is joined by Lightspeed partner Amber Yang for a week where agents turned up on both sides of the security line. A security researcher reported that one hacker used AI agents to break into more than a hundred company websites, including John Deere, by running a single prompt across thousands of agents at once. Amber explains what makes that approach work now, and how it differs from an agent swarm. Attackers also forced refrigeration systems at hundreds of French supermarkets to defrost, and a developer used Claude Code to strip preinstalled software off his parents’ TV. Then, Goodfire, a Lightspeed portfolio company, published an essay by researcher Nick Cammarata arguing that neural networks are far more readable than the field has assumed. Amber breaks down what he’s claiming, why it would matter for AI safety, and why it’s still a working hypothesis rather than a finding. DeepSeek and MiniMax both shipped efficiency updates the same week the US moved to remove DeepSeek from civilian federal systems, and Josh and Amber get into what that means for American companies choosing models on cost. They close on the personal agent race, including Claude Duet’s research preview, and Amber’s answer to what she’s looking for as an investor, which involves an unexpected parallel to the early oil business.Chapters:
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
Stay in touch: www.lsvp.com
X: https://twitter.com/lightspeedvp
LinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/
Instagram: https://www.instagram.com/lightspeed/
The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.
Meta and Google are making their next moves in the race for frontier AI, Apple is pushing deeper into the smart home, and SpaceX wants to become a major player in mobile. Claire and Josh discuss Meta’s Muse Wonder and Wonder Full models, including the company’s focus on active memory, persistent agents, and keeping its most capable model inside the Muse ecosystem. They also examine Google’s Gemini 4 Argon and its strengths across professional workflows, vision, coding, spatial reasoning, and long-running tasks.
Then, they look at Apple’s HomeView and why privacy, personalization, and ambient intelligence could give Apple an opening in the AI-powered home. Plus, they unpack what SpaceX would actually need to become a major U.S. wireless carrier and how a company that controls satellites, spectrum, and broadband could reshape the telecom market.
Chapters:
0:00 Welcome to Lightwork
1:04 Meta Muse Wonder vs. Google Gemini 4
5:00 Why AI Agents Need Memory and Focus
6:23 Gemini’s Advanced Reasoning and Task Automation
8:37 Apple HomeView: Siri Comes Home
11:33 AI Privacy and Smart Homes
16:52 SpaceX’s Plans To Be a Carrier
22:12 Zaria $5.2M Seed Round Led By Lightspeed
22:45 Zenzap Comes Out of Stealth
Stay in touch: www.lsvp.com
X: https://twitter.com/lightspeedvpLinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/Instagram: https://www.instagram.com/lightspeed/
The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.
Reflection co-founder Misha Laskin sits down with Claire Zau in Washington, D.C. to explain why AI labs need a presence in policy conversations and where the open weights debate stands right now. Misha compares closed and open models to renting and owning a home, and says companies that started by renting intelligence are now moving toward ownership. He describes a shift on AI gateways over the past six months from roughly 70 percent closed tokens to a majority of open ones.
They also cover some of Misha’s arguments: why China's labs have led on open models, why AI startups want control over cost, compute and data, and why that open source is the safer way to deploy AI. He makes the case through the history of strong encryption and Linus's law, and through the recent incident involving OpenAI and Hugging Face.
Stay in touch: www.lsvp.comX: / lightspeedvp LinkedIn: / lightspeed-venture-partners Instagram: https://www.instagram.com/lightspeed/...The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.Chapters: 00:00 Why AI Labs Need a Presence in DC01:15 The Open Weights Fight Explained02:09 Closed vs. Open Models: The Real Estate Analogy03:16 Why China Is Leading Open Models05:05 Why Developers and Startups Are Adopting Open Models06:45 AI Safety and the Open Source Debate08:19 Linus’s Law Applied to AI Safety09:26 Why Open Models Are Actually Safer10:44 Open Models in Practice: The Hugging Face IncidentvShifting From Renting AI to Owning It • Shifting From Renting AI to Owning It
Josh Machiz is joined by Shaurya Aggarwal, a partner on Lightspeed's growth team, for the week the West showed up to the open model fight. Reflection unveiled Beam on Monday, and Mistral unveiled Large 4, better known as Le Chonk, on Tuesday. Shaurya explains what open weights actually buy you, why both companies are competing on intelligence per dollar instead of chasing the frontier, and he breaks down mixture of experts in a way that makes the huge parameter counts finally make sense. Only about 5% of Beam wakes up for any given token. Lightspeed led Mistral's inception round and has invested in every round since, and co-led Reflection's Series A.Then DoorDash, which showed a delivery drone it designed itself plus an AI agent you can text to order. Shaurya's take is that the marketplace is the moat, and building its own agent is offense and defense at once against the assistants people already use. On that note, Instinct can now join your group chat, and Meta and Sierra announced an open standard for how personal agents work with businesses. Finally Oura, which called off its Nasdaq IPO hours before pricing despite heavy demand, and what that says about late stage growth investing. Plus updates from RobCo and Avarra. Lightspeed led RobCo's Series B, co-led its Series C and participated in this round, and led Avarra's seed and participated in its Series A.Chapters:00:00 Open Models, Drone Delivery, and AI Agents01:08 Mistral's Le Chonk Meme and Naming Story02:33 What Open-Weight Models Mean for the AI Race03:56 When to Choose Open Models Over Frontier AI05:30 Why Open Models Compete on Cost Efficiency06:35 Mixture of Experts Explained07:32 Is Reflection the DeepSeek of the West?07:58 Why DoorDash Is Building Its Own Delivery Drones09:38 DoorDash's Competitive Edge 12:28 DoorDash's Moat Against AI Agents14:34 Personal AI Agents in Group Chats16:08 How AI Agents Could Change Consumer Spending17:26 Meta, Sierra, and the Future of Personal AI Agents19:56 Oura Postpones Its IPO22:13 Lightspeed UniverseStay in touch: www.lsvp.comX: / lightspeedvp LinkedIn: / lightspeed-venture-partners Instagram: / lightspeedventurepartners The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.Reflection Beam, Mistral Le Chonk & the Drone That Delivers Your Chipotle | Lightwork • Reflection Beam, Mistral Le Chonk & the Dr...
Robotics had a big summer, and it is genuinely hard to tell who is doing what. Claire Zau met with Lightspeed partner Amber Yang in front of a whiteboard to draw the whole market on two axes. How general is the model, and how general is the body.
Yang explains what actually changed at the model layer, why robots cannot just be trained on the internet the way language models were, and why she thinks humanoids are not about to get cheap. Roughly 40 to 60 percent of a humanoid build is actuators, and that supply chain runs through China. Then she fills in the map. Dexterity, Tacta and Reflex Robotics building their own models for their own hardware. Ultra skipping training entirely and fine-tuning on Physical Intelligence. Skild and Physical Intelligence chasing one brain for many bodies. Sunday Robotics, 1X and Figure on the humanoid side. She also makes the case for what she calls DeployCos, companies that take someone else's model into places like a restaurant or a nursing home without the customer ever knowing what is under the hood. Her closing point is the one we believe worth sitting with. Robotics still has no shared benchmark, so every company grades its own homework, and in a field this tied to safety that gap may be an opportunity.
Stay in touch: www.lsvp.comX: https://twitter.com/lightspeedvpLinkedIn: https://www.linkedin.com/company/lightspeed-venture-partners/Instagram: https://www.instagram.com/lightspeed/The content here does not constitute tax, legal, business or investment advice or an offer to provide such advice, should not be construed as advocating the purchase or sale of any security or investment or a recommendation of any company, and is not an offer, or solicitation of an offer, for the purchase or sale of any security or investment product. The views expressed by our guests do not necessarily represent the views of Lightspeed. For more details please see lsvp.com/legal.