Lightspeed

"When you have a small set of people working on the most impactful technology in the world, they can't foresee all the consequences,” Reflection co-founder Misha Laskin told Claire Zau. He points to the OpenAI / Hugging Face incident. When closed models were too restricted to help with cyber defense, Hugging Face turned to an open model instead. "The bad actor argument is very theoretical." "In practice, we've actually seen the opposite. It was the open models that came in to help remediate the incident." @MishaLaskin on Lightwork

Vídeo

Reflection AI co-founder and CEO @MishaLaskin on why the economics of AI could change as adoption scales: “When you just leave home and you’re out of college, you typically go and rent an apartment. But as you grow… you might buy a home.” “In AI, the closed and open models… you can think about them as rental and ownership markets.” “Now, as organizations have adopted AI at scale… maybe they’re spending hundreds of millions of dollars on it, and rental no longer makes sense.” From our recent AI policy discussion in Washington, D.C.

Vídeo

Why do open models matter for America’s technological leadership? That was the focus of our fireside chat in Washington, D.C., introduced by Lightspeed Operating Partner Eric Traupe. @JoshuaTLevine, Director of Technology & Statecraft at @joinfai, highlighted the role of open research in American technological progress. @MishaLaskin, CEO and co-founder of @reflection_ai, discussed how openness allows researchers and engineers to investigate how models work and develop safeguards. The discussion was moderated by @ashleyrgold, Senior Tech Correspondent at @semafor, and brought together government and technology leaders. Thank you to everyone who joined us and shared their perspectives.

Vídeo

On Friday, the UK Government moved to ban non-compete clauses in UK employment contracts. A fantastic step for UK innovation. Our co-founder @ravi_lsvp explains why this matters for talent and the future of British tech: “Silicon Valley didn’t become Silicon Valley by making it slightly cheaper to hold people in place. It became Silicon Valley by letting them go and trusting in the collective ambition of brilliant minds. Britain can do the same.” The UK has world-class talent and exceptional companies, making it an even better place to build or join one. We’re proud to back founders who started in the UK and those expanding their teams here. More in The @ObserverUK.

Read the full piece here: observer.co.uk/news/business/…

Who’s building what in robotics? @theamberyang joins @clairejyz to map example companies by how general or specialized their models are, and whether they support one kind of robot or many. The chart uses examples from Amber’s discussion to illustrate her framework, not a complete view of the market. Full conversation from Lightwork below.

America needs competitive open models. @MishaLaskin, CEO and co-founder of @reflection_ai, argues that AI labs need a voice in D.C. to help policymakers understand the technology they’re regulating. He sees Chinese labs’ open releases as part of a competition for global adoption. American open models give developers another option for the technology they build on. For startups, the appeal is practical: they can fine-tune models on their own data and run them on infrastructure they control. Misha connects those capabilities to companies seeking greater independence as they build their products. On safety, he argues for “a distributed defense where you have many people, many contributors looking at a problem.” Drawing on cybersecurity, he makes the case that wider testing can help researchers identify and fix vulnerabilities. Misha shares his perspective on Lightwork with @clairejyz. Watch the full video below. 00:00 Why AI Labs Need a Presence in DC 01:15 The Open Weights Fight Explained 02:09 Closed vs. Open Models: The Real Estate Analogy 03:16 Why China Is Leading Open Models 05:05 Why Developers and Startups Are Adopting Open Models 06:45 AI Safety and the Open Source Debate 08:19 Linus’s Law Applied to AI Safety 09:26 Why Open Models Are Actually Safer 10:44 Open Models in Practice: The Hugging Face Incident

Vídeo

YouTube: youtube.com/watch?v=T1TSK7…

The economics of enterprise AI adoption no longer align. A new Forbes article dissects how @FireworksAI_HQ is helping companies manage AI costs without sacrificing performance. Roughly 10,000 enterprises currently use the platform to build specialized AI on open models, with more control over how those models run and what they cost. @ml_angelopoulos, Co-Founder and CEO of Arena, tells Forbes that many businesses “don’t even know how much they’re spending on every query.” We believe in the mission of Fireworks CEO and co-founder @lqiao and the entire team to make AI more efficient. As AI becomes more integrated into the enterprise, robust efficiency becomes essential. Read the full article here: t.co/IgW8xTs43M

Two new open-weight models in two days. Lightspeed portfolio companies Reflection and Mistral unveiled Beam and Large 4, all striving to bring greater control, customization, and efficiency to enterprise AI. More on Lightwork from @Machiz and guest co-host, Partner @AggShaurya: → DoorDash builds its own delivery drone and AI ordering agent. → Instinct brings agents into group chats as Meta and Sierra develop new rules for agent access. → Oura postpones its IPO despite reported strong demand. 00:00 Open Models, Drone Delivery, and AI Agents 01:08 Mistral's Le Chonk Meme and Naming Story 02:33 What Open-Weight Models Mean for the AI Race 03:56 When to Consider Open Models Over Frontier AI 05:30 Why Open Models Compete on Cost Efficiency 06:35 Mixture of Experts Explained 07:32 Is Reflection the DeepSeek of the West? 07:58 Why DoorDash Is Building Its Own Delivery Drones 09:38 DoorDash's Competitive Edge 12:28 DoorDash's Moat Against AI Agents 14:34 Personal AI Agents in Group Chats 16:08 How AI Agents Could Change Consumer Spending 17:26 Meta, Sierra, and the Future of Personal AI Agents 19:56 Oura Postpones Its IPO 22:13 Lightspeed Universe

Vídeo

YouTube: youtube.com/watch?v=KAdav9…

“It’s not rocket science…” Except this time, it is. Starship reached orbit for the first time, even after an engine issue put the attempt in doubt. Watching the setbacks and breakthroughs unfold live can make you feel like part of the team. @Machiz calls it “a building-in-public master class.” @clairejyz adds: “We are a little speck in the universe, and the fact that we’ve been able to accomplish breaking into orbit out of that is, I think, pretty inspiring.” Tune in to Lightwork for more!

Vídeo

We believe the AI boom is increasing venture capital’s importance to institutional investors. AI companies can reshape industries well before they go public, which may have implications for businesses across an institution’s portfolio. Working with founders gives our investors a view into how AI is being adopted and how business models are changing. We believe the institutions that back us should be able to draw on that perspective. Adil Haque has joined Lightspeed as Partner and Head of Capital Partnerships to help build those relationships. He’ll lead the team and work directly with our institutional partners to understand their priorities and connect them with our investors’ expertise. More on Adil’s appointment in @WSJ.

More on Adil’s new role at Lightspeed, via @WSJ: wsj.com/pro/venture-ca…

We believe the AI boom is increasing venture capital’s importance to institutional investors. AI companies can reshape industries well before they go public, which may have implications for businesses across an institution’s portfolio. Working with founders gives our investors a view into how AI is being adopted and how business models are changing. We believe the institutions that back us should be able to draw on that perspective. Adil Haque has joined Lightspeed as Partner and Head of Capital Partnerships to help build those relationships. He’ll lead the team and work directly with our institutional partners to understand their priorities and connect them with our investors’ expertise. More on Adil’s appointment in @WSJ.

More on Adil’s new role at Lightspeed, via @WSJ: wsj.com/pro/venture-ca…

We’re excited to welcome Adil Haque to Lightspeed as Partner and Head of Capital Partnerships. Adil brings more than two decades of experience at Apax and EQT. At Apax, he invested across venture, growth, buyout and hedge fund strategies before moving into investor relations. During his 12 years at EQT, he co-led the expansion of investor relationships across the Americas. He later led institutional product management globally, overseeing fundraising execution across the firm’s institutional funds. He’ll build and lead our Capital Partnerships team, bringing firsthand investing experience to conversations with our institutional partners about our investment approach and their long-term priorities. Meet Adil below.

Vídeo

“Software really isn’t about the code or the technology… It’s about the relationships.” Nilam (@beaconholdings Founder & CEO) argues that AI will reach small and mid-sized businesses through the software providers they already trust. @nilamg joined Lightspeed Partner @bsomaia on The Investment Memo to break down the thesis behind Beacon and how they’re operating. As he considered what kind of company he wanted to build, Nilam reflected on a recurring mission in his career: “I realized there was this common theme around making things fair for the people who otherwise are not given the same tools that I say that I was given,” he said. With that in mind, Nilam decided he could merge this passion and his vision for AI into Beacon, his company designed to expand AI capabilities for trusted software businesses across the country, from campgrounds to construction to transportation. He follows three key stages: finding and acquiring the right businesses, integrating them onto the Beacon platform, and building the AI products their customers need next. 00:00 Intro Nilam Ganenthiran 02:41 Why Nilam Left Investing to Build Beacon 05:51 What Is the Real Economy? 06:49 Why Small Businesses Adopt AI Slower 08:37 Why Beacon Needs to Exist Now 10:22 Finding and Buying Businesses With AI 12:12 Integrating and Running Acquired Companies 13:20 Growing Net Dollar Retention as the Hero Metric 14:03 Lessons From Instacart Applied to Beacon 17:25 How Big Can Beacon Get? 19:45 Beacon vs. Traditional Software Roll-Ups 25:27 What Happens to Founders After Acquisition 28:02 The Data Flywheel Across 40+ Companies 29:47 Change Management Is the Hard Part

Vídeo

We’re excited to double down for @arena’s $200M Series B. As AI models multiply, knowing how they actually perform in the real world is becoming increasingly important. Arena is also launching its Alignment Index, which measures how closely AI behavior aligns with human values in real-world settings, contributing to Arena's mission to measure AI through the experience of people who use it. We first partnered with the team at the seed. Since their Series A, Arena has exceeded $100M in annualized revenue, with millions of people helping evaluate frontier models through real-world use. That growth reinforces our original investment thesis that the more models there are, the more valuable independent signals about their performance become. Congratulations to co-founders @ml_angelopoulos, @infwinston, and @istoica05. As the model race gets more crowded, Arena’s work has been getting even more interesting. CC: @buckymoore, @jamesalcorn94, @ravi_lsvp, @prathikhosalli

Learn more about Arena here: arena.ai/blog/series-b

Ahmet Golshan, NVIDIA’s Senior Director of AI Software, on why the next generation of AI security needs to be built for an agent-native world: “True agent-native identity management is something that needs work.” “Just passing human permissions to agents is not right.” “We’ve seen the industry try to forklift traditional things and see if it applies.” “I think we’re at that inflection point where we’re seeing the breakages of what happens in an agent-native world.” The greenfield opportunity: identity, privacy, security tooling, and observability built specifically for agents. @a_golshan @clairejyz

Vídeo

Choosing an AI assistant is starting to feel a little like choosing a pet. @clairejyz calls it “the Tamagotchification of tech.” OpenAI unveiled Dots, always-on agents with their own cloud computers that can use browsers and connected apps to complete tasks. As assistants learn your preferences and get things done, personality could become another reason to keep the same one around. @Machiz asks, “Do you want the smartest intelligence or do you want the one that makes you want to give it a hug?” Watch the full segment below.

Vídeo

The humanoid cost curve has a big bottleneck: actuators. Lightspeed Partner Amber Yang on why the hardware side of robotics may not commoditize as quickly as people expect: "40-60% of a humanoid's build is actually spent on actuators." "Today, we are almost completely relying on China for actuator supply chain." "All these actuators are built with rare earth magnets that we're also completely dependent on China's supply chain for." "It's not necessarily a fair claim to say that humanoids as a hardware embodiment will become commoditized." @theamberyang, Partner at Lightspeed, on Lightwork:

Vídeo

What if AI could handle the paperwork you’ve been putting off? “Being able to file a passport online and have it be done through a chatbot just feels like the future,” says @Machiz. That’s part of the longer-term ambition for t.co/BBp2rxokaB, which adds an AI search-and-answer layer over roughly 29,000 government websites. “That is the value of AI: it’s able to take a lot of information and make it navigable and searchable,” shares @clairejyz. Meanwhile, AI leaders signed a voluntary safety accord at the White House as Congress weighs its next steps on AI oversight. Watch the full segment below.

Vídeo

We’re excited to lead @gardensdevs' $35M Series B. As more of the world becomes automated, Gardens is betting that shared experiences between real people will matter even more. Co-founders @GardensChris, Lexie Dostal, and Stephen Bell are building a social role-playing world they call a “strandbox,” shaped by the people who pass through it. Players’ choices live on long after they leave, influencing the experiences of people they may never meet. The game was inspired by the team’s memories of online gaming and the magic of sharing stories with strangers. We first invested in Gardens at their Series A in 2023, and are delighted to double down. Congratulations to the team on this new chapter! @bsomaia

Read more about Gardens Interactive here: bloomberg.com/news/articles/…

Robots are different from LLMs because there is no giant corpus of internet data to train them on. "With robotics companies, they've had to rely on a number of data collection techniques to really scale up their training." Amber Yang explains why in-context learning is becoming so important: "In-context learning is fascinating because in real time you're observing when the robot does well versus badly." "That's potentially the highest quality type of data because it's the robot-specific embodiment." @theamberyang, Partner at Lightspeed, on Lightwork:

Vídeo

Healthcare's next chapter could be written in software. The @TIME World's Top HealthTech Companies of 2026 list is here, recognizing companies across the globe that are using digital technologies aiming to make healthcare more effective and accessible. We’re thrilled that four current and former Lightspeed-backed portfolio companies made this year’s list. Congratulations to: → @AbridgeHQ → @Innovaccer* → @Neko → @MDCloneHQ Lightspeed is proud to back teams working to help build a healthier future. You can see the full list here: t.co/mLbB1AXDfl *Innovaccer is a former portfolio company of Lightspeed India Partners (”LSIP”), a separate and distinct investment advisory entity from Lightspeed Management company, L.L.C. (”LSVP”). LSVP and LSIP are not a unitary enterprise. Lightspeed India has exited its position in Innovaccer and no longer holds an interest in the company.

We’re thrilled to continue backing Avarra for its $17M Series A. While you can give a sales rep a playbook, you can’t give them hundreds of hours of learned experience. Avarra captures top performers’ know-how in AI avatars that help GTM teams close knowledge gaps and build their skills. Reps can practice complex sales scenarios, prepare for calls, and get support during live deals. The same avatars can work directly with buyers. Updates to the playbook carry through to both training and customer conversations. Avarra reports more than 10,000 AI coaching and role-play sessions each week. We led Avarra’s seed round in 2023 and are excited to continue backing co-founders David Knight, Matt Huang, Scot McIntosh, and Christopher Iezzoni. CC: @guruchahal, Venture Partner @arifj

Learn about Avarra here: avarra.ai/blog/seriesa

How do you know which robots are ready for the job? Robotics companies test their models on different tasks and datasets. Lightspeed Partner @theamberyang sees a need for better ways to compare them, especially when customers are making decisions about safety. She maps the robotics market by how general or specialized companies’ models and hardware are. As models become more adaptable, Amber sees an opportunity for companies that specialize in deployment. She calls them “DeployCos”: businesses that could pair existing models with different kinds of embodiments to handle work in places like factories or restaurants. 0:00 Why Older Robotics Companies Struggled 0:40 Mapping The Robotics Landscape Across Two Axes 1:16 Model Layer Breakthroughs From VLAs To World Action Models 4:20 One Brain Many Bodies And The Value Of Specific Models 5:27 In-Context Learning And How Robotics Companies Collect Training Data 7:40 Hardware Developments And Dependence On China 9:24 Specific Model Specific Embodiment Quadrant 13:05 The Deploy Co Angle And Open Router For Robotics 13:55 General Model Multiple Bodies Quadrant 15:50 General Model Specific Embodiment Quadrant 17:24 Where Things May Be Commoditized And Infrastructure Opportunities 20:45 Robotics Observability And Evaluation 22:33 Learning From Autonomous Vehicles @clairejyz

Vídeo

@theamberyang YouTube: youtu.be/edCGAXnBcBI

@theamberyang Spotify: open.spotify.com/episode/5kYRAq…

Congrats to @reflection_ai on the launch of Beam: Reflection’s first open-weight model. 501B total parameters, 23B active, built for coding, reasoning, and agentic workloads. We believe this is a meaningful step towards more accessible AI for enterprises, public sector, and developers. We’re incredibly proud to support @MishaLaskin, @real_ioannis and the team. CC: @ravirajjain @JamesAlcorn94

Congrats to @reflection_ai on the launch of Beam: Reflection’s first open-weight model. 501B total parameters, 23B active, built for coding, reasoning, and agentic workloads. We believe this is a meaningful step towards more accessible AI for enterprises, public sector, and developers. We’re incredibly proud to support @MishaLaskin, @real_ioannis and the team.

We hosted a fireside chat last week to break down the value of open frontier models in America. Operating Partner Eric Taupe opened the conversation before introducing the speakers, including @MishaLaskin, the CEO and co-founder of @reflection_ai. Misha highlighted the significance of open research, noting that a larger scientific community enables more people to understand breakthroughs and mitigate concerns. Reflection's open models reflect this philosophy: they aim for AI to become the foundation on which everyone can build the next era of progress. Fellow speaker @JoshuaTLevine, the Director of Technology & Statecraft at @joinfai, connected this concept to the technical evolution in the US, which “has been defined by this fusion of being able to build in the open, set standards, and set the foundational levels of the tech stack.” FAI's work aims to forge a more perfect union between innovation and the American republic. @ashleyrgold, Senior Tech Correspondent at @semafor, moderated the discussion. Attendees included policymakers, tech leaders, and congressional staff. Thank you to Misha, Joshua, and Ashley for leading an insightful evening, and to all attendees for joining us.

Today is the last day to submit nominations for the 2026-2027 Cyber 60 list. Cyber 60 is an annual report presented by Lightspeed, in partnership with @FortuneMagazine and sponsor @AWScloud, spotlighting who we believe to be 60 of the most innovative and fastest-growing companies in cybersecurity. Today, AI continues to reshape the rules of cybersecurity as more startups answer the opportunity. Nominate your company today: t.co/JzBrTE2hGc

When robots automate the ordinary, humans have more time to do the extraordinary. Congratulations to @Robco_Robotics for reaching a $1B+ valuation. 1,000+ robots deployed. 3.5M+ operating hours. And a new autonomous industrial robot targeting launch in March. Lightspeed is proud to have been RobCo’s partner since leading the company’s Series B in 2024 and Series C earlier this year. #RobCoIsAUnicorn #FactoryFloorAI #PhysicalAI t.co/5zOhyO7dps

Why does AI need so much more energy than the human brain? "Most of the energy in a computing system goes into moving information around." "The computer from 1945 operates very similarly to how computers operate today. It has memory on the outside, some kind of computing, and you move bits back and forth." "We built computers to be fast. This computer is twice the speed of that computer; that's how we sell computers. But it doesn't contemplate energy efficiency, and that's what we're changing." @NaveenGRao, CEO of @unconvAI, at the @theallinpod Summit:

Vídeo

AI could radically change how people access government resources. The US is starting with a chatbot. This week on Lightwork: → t.co/BBp2rxokaB is now live. Designed as a front door to the federal government, the chatbot aims to make information easier to find across 29,000+ government websites. → Leaders across technology and government gathered for lunch at the White House to discuss the state of AI and what “Superintelligence” could mean for the US public. → OpenAI launched over twenty features at DevDay, including a system of “always-on” agents called Dots. Is consumer tech headed toward personal background workers? → SpaceX’s Starship reached orbit on Flight 14 and deployed its first operational payload, twenty-six Starlink V3 satellites. Full video below. @clairejyz @Machiz

Vídeo

YouTube: youtube.com/watch?v=fytn9z…

YouTube: youtube.com/watch?v=fytn9z…

Amazon generated $68.6 billion in advertising revenue in 2025, much of that from placements where shoppers browse and compare. What happens to that business when an agent does the shopping instead of a person? Part of Amazon’s advertising business reaches shoppers as they browse and compare products. If agents handle more of that comparison, shoppers may see fewer sponsored placements. That could change where brands pay to reach them. More on Lightwork: t.co/s08uCUCbtA

AI just got an audit performance review. @ModusAudit just launched Financial Audit Bench, putting 11 frontier models through audit workpaper tasks using synthetic data and 1,100+ hours of auditor input across development and review. One striking finding is that the top model passed 69.17% of individual runs, but passed all eight attempts on only 44.4% of tasks. We invested in Modus earlier this year, backing their mission to deliver higher-quality audits in a fraction of the time. This project presents the first release from Modus Labs, a team that brings leading PhDs and Applied AI Engineers together with some of the top minds in public accounting. Congrats @Arushj01, @PranavAPillai, @vinaykasat_, and the entire team. CC: Lightspeed Partners @jmover, Isaac Kim, Amish Desai, Alexa Boghosian, Steven Munsie

Read more in our blog: lsvp.com/stories/introd…

"The mentality of security has to change completely.” Congratulations to @Cyera CEO Yotam Segev on being named to the TIME100 Next 2026, commemorating some of the world's most influential rising stars. time.com/collection/tim…

Census’s August 2026 estimates put e-commerce at 17.1% of U.S. retail sales, up from 0.6% in late 1999. Agent-driven shopping asks people to hand over more of the purchasing decision. Reordering a familiar item may be an easier starting point than delegating a purchase that depends on fit or taste. That could make adoption uneven across categories, even as the technology improves. More on Lightwork: t.co/s08uCUCbtA

More than half of U.S. adults who have never used AI chatbots cite concerns about their personal data as a major reason. For shopping agents, that raises a practical question: what information can an assistant access, and which purchases still need your approval? “We've all given so much data to Meta, I feel like people are concerned about giving more of their data to the company.” @clairejyz on Lightwork.

Ande estimates companies spend $325 billion a year on client dinners, team outings, and sporting events. @keepingittlo, CEO of @weareande, on who ends up paying for it: "People are like, I don't have a corporate card, so I'm putting it on my personal card. Or I spent more than my limit. And then it becomes the finance team's job to reconcile this." From his appearance on Lightwork with @clairejyz and @Machiz

Vídeo

What happens if AI gets 1,000x cheaper? Jevons paradox says demand may rise even faster than the cost falls. "If you make something half the price, you'll potentially consume more than 2x." "If we disrupt AI by 1,000x, I think this will create the largest market humanity has ever seen." Naveen thinks this might create the largest market that humanity has ever seen. @NaveenGRao, CEO of @unconvAI, at the @theallinpod Summit:

Vídeo

Nominations are now open for the 4th annual Cyber 60 list. Cyber 60 is an annual report presented by Lightspeed, in partnership with @FortuneMagazine and sponsor @AWScloud, spotlighting some of the most innovative and fastest-growing companies in cybersecurity. The list recognizes teams building to advance next-gen infrastructure, AI security, and how we approach threat detection. 60 teams across early, mid, and late stages. Nominate your company today. Applications close on October 4th: t.co/JzBrTE2PvK

Healthcare founders build inside a system shaped by policy, reimbursement, and incentives. 6 takeaways from Venture Doctors Episode 1 with Jacob Shiff, Chief AI & Technology Officer at the CMS Innovation Center (@CMSGov) ↓

Healthcare founders build inside a system shaped by policy, reimbursement, and incentives. 6 takeaways from Venture Doctors Episode 1 with Jacob Shiff, Chief AI & Technology Officer at the CMS Innovation Center (@CMSGov) ↓

Hjalmar Nilsonne (@HNilsonne), founder and CEO of Neko Health, on why the company built its technology in-house. “We just didn't see that you were going to produce a lot of MRI machines and then you would solve preventive healthcare for 7 billion people.” “When you think about what's possible with AI, when you can cross-correlate hundreds of thousands or billions of different features into a better understanding…” “If you can have a different approach to collecting and tracking your health over time, this will be both affordable and highly, highly precise and valuable.” From Lightwork

Vídeo

Hjalmar Nilsonne (@HNilsonne), founder and CEO of Neko Health, walks through what happens during a Neko Health scan. “The first part is the diagnostic part.” “We map your skin. We map your heart. We map your arteries. We do a blood draw.” Then, you sit down with a prevention-focused clinician to go through all of the results. “You have 20 to 25 minutes of undisturbed attention to help make sure that you walk away with the most important insights you need to know about your health.” From Lightwork

Vídeo

Most retail innovation has focused on the consumer. @faire_wholesale has spent years working on how brands and independent retailers find each other and do business. Today, Faire’s platform is on pace to facilitate more than $4.5B in merchandise purchases annually. The company is also exploring fulfillment, where its scale could meaningfully lower shipping costs for independent retailers. Proud to have backed the team since 2018. Read more in @WSJ on what could come next. CC: James Ephrati, Board Partner @ataussig

Read more: wsj.com/business/retai…

The newest AI model can’t text back. TypeSafe released Jev, their “system one” model that doesn’t respond with paragraphs, only probabilities. While it can still be wrong, it can’t wander outside the answer schema the way a chat-based LLM can. Will frontier labs take notice and start building their own? More on Lightwork this week: → Amazon bans Meta Muse from their website to guardrail its $68B ads business, while storefront-style platforms like Shopify welcome agent use for traffic. → Skild AI trained its robots on 140 years of simulated self-play soccer, showing that physical self-play could enable robots to far exceed human capability. Full video below. 00:00 Welcome to Lightwork 01:07 Jev: The AI Model That Doesn't Chat 06:39 Real-World Uses for Jev 9:56 Amazon Blocks Meta's Muse 13:23 Why Shopify Is Embracing AI Shopping Agents 21:22 Why Meta's Muse Took Off 23:07 Muse's Safety & Privacy Approach 26:15 Behind the Scenes at Meta Connect 27:24 Skild AI: Physical Self-Play @clairejyz @Machiz

Vídeo

YouTube: youtube.com/watch?v=mIPbYs…

@Neko @HNilsonne @eldsjal Spotify: open.spotify.com/episode/2LnNM4…

YouTube: youtube.com/watch?v=XYITQZ…

Congratulations to @Neko Health on opening their first US clinic! With 8 locations already open across the U.K. and Sweden, today marks the first time Neko Health is made available in the US. From a DM conversation to an international brand. We’re proud to be backing @HNilsonne, @eldsjal, and the entire Neko Health team.

The future of healthcare is shaped by our reward systems. For the first episode of Venture Doctors, hosts and Lightspeed Partners @BrentonFargnoli and @GalymImanbayev sit down with Jacob Shiff, Chief AI & Technology Officer at @CMSGov. They unpacked CMS’ new ACCESS model, which ties payment to measurable patient outcomes rather than the technology itself. The three explored how AI could expand access to care and why aligning financial incentives with patient health may be essential to realizing that potential. They also discussed what this shift could mean for the future of healthcare and the founders building it. You can watch the full conversation at the link below. 0:00 Welcome to Venture Doctors 0:47 Introducing Jacob Shiff 1:25 Jacob Shiff's Path from Founder to the Government 3:29 The Listening Tour 7:51 The Access Model and How It Works For Patients 15:40 Expanding Beyond Medicare 19:17 AI Regulation and The Tempo Pilot 23:18 Government Service and Startup Speed At Scale 26:14 Advice for Founders Working with the Government 27:24 Underhyped Opportunities In AI And Healthcare 28:45 A Belief Most People Disagree With 29:57 What Jacob Is Digging Into Next

Vídeo

YouTube: youtube.com/watch?v=ax7cqB…

How do you build systems and agents that let one marketer produce the work of a hundred? We are excited to co-host a one-day hackathon with @Profound and @Google on Saturday, October 3. 40 builders. One day. $40,000 in available prizes. The prompt: Find a marketing process that's inhuman in scope or scale, and ship a system or agent that runs it. Profound will offer an interview to the winners, judged by a panel of marketing leaders from Google, Plaid, Kleiner Perkins and more. Apply here today: t.co/QXyrG9uLc3 CC: Eleanor Hooker, @MichaelC13191, @sach_pat, @elijahconniff

Vídeo

A month ago, S1 showed it could learn a new task from a single video. Now it can post-train through self-play without new demonstrations. Another step toward robots that can keep learning beyond human-provided data. Incredible progress from @SkildAI, @deepakpathak and @gupta_abhinav_

Two doctors walk into a venture capital firm. Now they have a show. Introducing Venture Doctors, a new show from Lightspeed exploring the intersection of medicine, technology, and venture capital. Hosted by physicians-turned-investors @BrentonFargnoli and @GalymImanbayev. To kick things off, they answered some rapid-fire questions: “Most overused word in healthcare venture?” Brenton: “AI.” Galym: “Agent-something.” First episode drops tomorrow at 10am ET.

Vídeo

Two doctors walk into a venture capital firm. Now they have a show. Introducing Venture Doctors, a new show from Lightspeed exploring the intersection of medicine, technology, and venture capital. Hosted by physicians-turned-investors Dr. Brenton Fargnoli and @GalymImanbayev. To kick things off, they answered some rapid-fire questions: “Most overused word in healthcare venture?” Brenton: “AI.” Galym: “Agent-something.” First episode drops tomorrow at 10am ET.

Vídeo

Two doctors walk into a venture capital firm. Now they have a show. Introducing Venture Doctors, a new show from Lightspeed exploring the intersection of medicine, technology, and venture capital. Hosted by physicians-turned-investors Dr. Brenton Fargnoli and @GalymImanbayev. To kick things off, they answered some rapid-fire questions: “Most overused word in healthcare venture?” Brenton: “AI.” Galym: “Agent-something.” First episode drops tomorrow at 10am ET.

The hardest AI data problems increasingly demand both human expertise and AI. @SnorkelAI pairs expert professionals with specialized models and agents to build the complex data and environments frontier systems need. Congratulations to @ajratner and the team on the Series E! CC: @ravirajjain, @ravi_lsvp, @whoisnnamdi

What actually happens when you get your spoon bent? Claire explains her take on Bending Spoons' playbook and what it could mean for the unicorns of 2021: "Acquire these strong brands that have stalled growth, replace most of the team, rebuild them with a centralized platform, raise prices." "They actually want to hold them forever. It's not really this flipping model." "If you are a 2021 unicorn, there are two very clear comps that you already have with Airtable and Miro. It is a very blunt message for founders." @clairejyz

Vídeo

We're excited to have co-led @weareande's Seed and Series A financing rounds as they emerge from stealth. Ande is building the AI-native network for corporate entertainment, connecting the enterprises that book experiences with the vendors that serve them to make corporate entertainment feel effortless. By bringing booking, approvals, contracts, payments, and reporting into one network, Ande is adding structure to a category — corporate entertainment — that has stayed scattered and largely invisible, the way travel was before it became a managed spend category. The company is AI-native internally, too. Founder Lohit Sarma says 95% of Ande’s code is now written by AI. Congratulations to @keepingittlo and the team on the raise and the public launch! Cc: Venture Partner @arifj

Read more about the raise here: theinformation.com/newsletters/ai…

While corporate travel has been centralized for decades, corporate entertainment is still comparatively scattered. On Lightwork, Lohit Sarma shared how he approached that problem at @weareande: start with supply. He spent the last two and a half years building the network from that side first, including going door to door with hospitality groups before scaling the enterprise side. Corporate entertainment still runs on relationships. AI can handle things like availability, pricing, and multi-player workflows, but Lohit draws the line at negotiation. As he puts it, “one bad event is one lost relationship that they can’t recover from.” Tune into the full episode below! Chapters: 0:00 Ande Comes Out Of Stealth 2:17 Why Corporate Entertainment Has Stayed Fragmented 4:51 Lohit Sarma's Background (Lifion by ADP) 9:51 What CFOs Revealed About Fraud And Offshore Teams 10:47 Building The Supply Side First 16:44 Ande's Enterprise Customers And Rollout Strategy 18:40 How AI Is Built Into Ande's Ethos 22:58 The Future Of Ande 26:11 Hiring A Team To Build A Category 29:38 What Can Break At Ten Times The Size 31:50 Celebrating The Launch And Where To Find Ande @keepingittlo @clairejyz @machiz

Vídeo

YouTube: youtube.com/watch?v=A6pTiZ…

@weareande Spotify: open.spotify.com/episode/0MaLjL…

We're excited to have co-led @Ande’s Seed and Series A financing rounds as they emerge from stealth. Ande is building the AI-native network for corporate entertainment, connecting the enterprises that book experiences with the vendors that serve them to make corporate entertainment feel effortless. By bringing booking, approvals, contracts, payments, and reporting into one network, Ande is adding structure to a category — corporate entertainment — that has stayed scattered and largely invisible, the way travel was before it became a managed spend category. The company is AI-native internally, too. Founder Lohit Sarma says 95% of Ande’s code is now written by AI. Congratulations to @keepingittlo and the team on the raise and the public launch! Cc: Venture Partner @arifj

Read more about the raise here: theinformation.com/newsletters/ai…

“My Vampire System,” a vampire-meets-science-fiction series, has made over $90 million as of September 2026. Pocket FM cofounder and CEO @RohanNayak2: “If someone in Hollywood looked at it, they wouldn’t have greenlit that. They would never have greenlit that. But users should vote for what they like. Users should tell us, or anyone, what a hit is.” From Lightwork

Vídeo

You can check whether AI got a math problem right but how do you check whether it made a good ad? Profound CEO on why building a marketing benchmark means working with thousands of marketers: "When there's a verifiable correct response, AI will be able to attack that vertical almost autonomously." "When we enter into these non-verifiable domains, it is much more dependent on human review and human approval." "Building the benchmark has required us collaborating with hundreds of our customers and thousands of marketers to build a robust benchmark that captures the consensus: is this good marketing? Is this bad marketing?" @thejamescad on Lightwork

Vídeo

Streaming services democratized entertainment consumption, the creator economy democratized distribution, and AI could do the same for high-quality production. @RohanNayak2, Co-Founder and CEO of @PocketFM_app, gave us the scoop on the company’s thesis behind their AI-driven audio drama platform. Pocket FM doesn’t rely on general-purpose LLMs alone. Rohan explains that Sherpa, its multi-agent fiction-writing system built on custom post-trained models, focuses on long-horizon narrative memory, emotionally convincing prose, and long-form planning. Rohan also gets into the harder problem that comes with cheaper creation. More content means Pocket has to get better at identifying the stories people actually want to keep listening to. Pocket’s fiction models are trained on billions of minutes of engagement data, while listener behavior helps determine which stories reach larger audiences. The team is now driving their newest product, Pocket Saga, an app that converts the best audio dramas into AI-animated vertical video series. Rohan sees AI as a tool that creators can use to execute on their ideas. He believes: “For the first time, literally anyone in the world can create studio-quality entertainment at their home.” Watch the full episode with Rohan on Lightwork at the link in the comments.

Vídeo

YouTube: youtu.be/9Z1lnlmVBN0

Pacing the Frontier: AI capability seems to be compounding faster than the industry's understanding of how to control it. Some tech leaders are advocating for pacing AI. Others are pushing against it. More on Lightwork this week: → @suno launched a new model in partnership with record companies Warner Music Group and BMG. → Licensing has split off into two camps: “open studio” vs. “walled garden,” either allowing or blocking download access for AI-generated audio. The jury’s still out on the winner. → Bending Spoons’ $1.355B acquisition of Miro sparks a conversation about the future of collaborative “pandemic era software”. Full video below. @clairejyz @Machiz

Vídeo

YouTube: youtu.be/WScxb5Ql8Y0

Saving time and money is serious business. Just ask @tryramp. Congrats @eglyman, @karimatiyeh and team on the UK launch!

We’re doubling down on @raindrop_ai in its Series A, bringing the company’s total funding to $50M. Agents can pass a team’s test cases and still fail in ways no one anticipated. What happens in production needs to inform how teams test their next release. Raindrop Simulations, now in research preview, replays real production traffic and existing test cases against a proposed agent change, then looks for unexpected behavior in the results. Teams can catch issues before deploying the change. We first backed @alexisgauba, @snarkyzk, and @benhylak at seed, and we look forward to supporting the team as they continue building “the sentry for AI agents.” Congratulations to the entire team on the raise!

Read more about Raindrop here: axios.com/pro/enterprise…

“ERP used to be a product companies would take on at Series B or later, when they had the available resources.” “DualEntry is seeing much smaller companies, even seed companies, adopt its product to get better insights and the ability to segment their books.” @seisler1 on how AI upended the ERP category. t.co/DHpJt7N52h

Lightspeed Partner Bucky Moore on the blurry line between personal and enterprise AI assistants: “It’s hard for me to see a world in which there will be a perfect bifurcation between working and personal assistance, because that’s just not how we live.” “Once I connect something that knows everything about my personal life, how do I know what’s flowing into the space at which it understands my work?” “There are some very interesting permissioning and segmentation problems that you can probably think about solving.” @buckymoore on Lightwork

Vídeo

Nearly every marketer is talking about how their brand shows up in AI. Today, @profound is the fastest-growing AI marketing platform, used by more than 1,000 enterprise brands including one third of the Fortune 100. @thejamescad (CEO & Co-Founder of Profound) joins @clairejyz to break down the emerging role of the marketing engineer, Profound’s work to build a comprehensive marketing benchmark, and why he believes the future of marketing will be AI-driven and human-supervised.

When your AI agent goes down, your company's reputation could go with it. Temporal co-founder on why reliability can't be something AI teams figure out later: "As long as it's working, no one cares. But the moment there is a problem, people realize how important that backbone is." "Once these systems are down, it causes so much pain to the customers. It erodes the trust in whatever product, whatever brand you are trying to build." "I see it shifting where people will be thinking about what's the reliability backbone you want to have in place before you build these AI systems on top of it." @SamarAtTemporal @temporalio

Vídeo

Dual Entry CEO @santinestea on doing an ERP migration while flying a plane: “We got ourselves a tiny Starlink. We put it out in the window.” “I opened up the laptop from the air and got them end-to-end through the entire migration. All their banks connected, all their integrations live, the books reconciled down to the penny before we landed.” With @seisler1 and @clairejyz From Lightwork

Vídeo