Frank Petterson

Dudu Lasry

TL;DR: We’re pivoting - from building an AI SRE to building Cursor for pharma and medtech companies. Here’s what we’ve learned. The past few months have been a roller coaster for Topaz T. and me, and I didn’t post as much as I’d like We started Vespper (YC F24) from a personal pain: as a developer, I was thrown into endless on-call shifts: "Customer X is experiencing issues, what’s going on?" When you’re on call for dozens of services, failure points are everywhere. As a result, I really appreciated teammates who jumped in with helpful insights: - "RabbitMQ is exploding" - "Add some-feature-id to ALLOWED_FEATURE_IDS" When GPT-4 arrived I wondered: what if a digital on-call teammate did that - scan gazillion logs and dashboards and surface the issue? That spark led me to meet Topaz, start a company, and join Y Combinator. YC was wild: hackathons in an Airbnb, bug-fixing, customer calls, and learning from people like Sam Altman, Tony Xu, Kyle Vogt, and Parker Conrad. We grew fast, but after demo day we felt that something is not working. The product wasn’t creating enough value. I've realized two things during that period. First, in B2B the product must deliver real, tangible value (save or make money). Second is momentum. I found a great post by Erez Druk called ‘Guess It Until You Make It’, and after speaking with Erez, I learned we should decide and not wait. So we made the hard call to pivot. Before Vespper I worked at Viz.ai and saw how much work regulatory, clinical, and quality teams do just to get to market - and stay there. After months talking with experts in pharma, medtech, and biotech, the picture became clear: - Preparing submissions (IND, BLA, NDA, 510(k), CSR, narratives) is painfully manual. Even "simple" drafts take weeks/months. - Timelines are extreme. INDs take months. NDAs/BLAs are 4–5x longer. For late-stage assets (e.g drugs), delay costs ~1ドルM/day (!). - External search is fragmented. FDA/EMA workflows are slow and scattered. We looked at the current tools and saw more gaps: - Legacy tools are rigid and outdated. Hard to adapt to real data and evolving workflows. - Current AI tools collapse on massive corpora needing retrieval, grounding, and traceability. A submission package often contains thousands of pages (~2 GB and more). That’s the opportunity we’re tackling. In the past weeks we’ve built an AI-powered workspace for regulatory and quality teams: - Generate first drafts of complex docs directly from data (IND, BLA, CSR, narratives, 510(k)). - Search across internal files and external databases (FDA 510(k), MAUDE, etc.). - Use a ChatGPT-grade copilot for anything regulatory. It’s early, but experts already report meaningful time and cost savings versus generic AI tools. We’re excited to share more soon. Next month we’ll be at a few conferences - details to come. Whether you’re in life sciences curious about our product, or just want to talk AI and startups, feel free to DM me! 😊

TL;DR: We’re pivoting - from building an AI SRE to building Cursor for pharma and medtech companies. Here’s what we’ve learned. The past few months have been a roller coaster for Topaz T. and me, and I didn’t post as much as I’d like We started Vespper (YC F24) from a personal pain: as a developer, I was thrown into endless on-call shifts: "Customer X is experiencing issues, what’s going on?" When you’re on call for dozens of services, failure points are everywhere. As a result, I really appreciated teammates who jumped in with helpful insights: - "RabbitMQ is exploding" - "Add some-feature-id to ALLOWED_FEATURE_IDS" When GPT-4 arrived I wondered: what if a digital on-call teammate did that - scan gazillion logs and dashboards and surface the issue? That spark led me to meet Topaz, start a company, and join Y Combinator. YC was wild: hackathons in an Airbnb, bug-fixing, customer calls, and learning from people like Sam Altman, Tony Xu, Kyle Vogt, and Parker Conrad. We grew fast, but after demo day we felt that something is not working. The product wasn’t creating enough value. I've realized two things during that period. First, in B2B the product must deliver real, tangible value (save or make money). Second is momentum. I found a great post by Erez Druk called ‘Guess It Until You Make It’, and after speaking with Erez, I learned we should decide and not wait. So we made the hard call to pivot. Before Vespper I worked at Viz.ai and saw how much work regulatory, clinical, and quality teams do just to get to market - and stay there. After months talking with experts in pharma, medtech, and biotech, the picture became clear: - Preparing submissions (IND, BLA, NDA, 510(k), CSR, narratives) is painfully manual. Even "simple" drafts take weeks/months. - Timelines are extreme. INDs take months. NDAs/BLAs are 4–5x longer. For late-stage assets (e.g drugs), delay costs ~1ドルM/day (!). - External search is fragmented. FDA/EMA workflows are slow and scattered. We looked at the current tools and saw more gaps: - Legacy tools are rigid and outdated. Hard to adapt to real data and evolving workflows. - Current AI tools collapse on massive corpora needing retrieval, grounding, and traceability. A submission package often contains thousands of pages (~2 GB and more). That’s the opportunity we’re tackling. In the past weeks we’ve built an AI-powered workspace for regulatory and quality teams: - Generate first drafts of complex docs directly from data (IND, BLA, CSR, narratives, 510(k)). - Search across internal files and external databases (FDA 510(k), MAUDE, etc.). - Use a ChatGPT-grade copilot for anything regulatory. It’s early, but experts already report meaningful time and cost savings versus generic AI tools. We’re excited to share more soon. Next month we’ll be at a few conferences - details to come. Whether you’re in life sciences curious about our product, or just want to talk AI and startups, feel free to DM me! 😊

Kelly Stonelake

For 15 years, I was trusted to lead Meta's most important teams and projects. I didn’t expect my career to end when I took a stand against a racist video game, but it’s true. When I was handed the opportunity to lead the company through the expansion of Horizon, I was thrilled. However, I was horrified to walk into a room of men wallpapering over rampant racism, a product riddled with performance issues, and violations of public policy. Worst of all, the victims were predominantly children. When I raised the problem, I became the problem – a pattern that silences women everywhere, every day. I’ve been sexually assaulted by a boss on a business trip. I’ve been denied promotions because acknowledging my success meant acknowledging men’s failures. I’ve been told to act "less smart," I’ve been retaliated against for doing my job. Some of you are shocked because you had no idea these things happened – some of you are shocked because you had no idea we could be honest about it. My privilege means I have the responsibility to speak up, but the discrimination and hostile work environment I experienced happens at every level, in every industry. It’s exponentially worse for women of color. It creates bad business outcomes that disproportionately harm those we should be most eager to protect, it widens the wealth gap, it puts lives at risk, and it’s against the law. It cost me my career, it almost cost my life. I’d been crotch grabbed, screamed at, told to have sex with my boss for a promotion. I survived all of it. Nothing broke me like a job where I had to tell powerful men "no." When tech companies push marginalized leaders out, they build dangerous products. This isn’t just an ethical problem. It’s a governance failure. It puts employees, shareholders, and users everywhere - especially those most in need of protection - at risk. When Meta's leadership dismissed and excluded me, they weren't just sidelining women, they were prioritizing power over people. They were putting their growth before the good. This cycle repeats itself across the industry: Women, minorities, neurodivergent people, raise ethical red flags about product safety They face retaliation and exclusion for speaking truth to power Harm then falls disproportionately on the most vulnerable users In the near-term, tech companies must strengthen protections for ethical whistleblowers, make promotion practices transparent, recommit to diversity programs and attach their success to executive compensation, and hardwire compliance into the product development process. Where Mark Zuckerberg rants about companies needing to be more masculine, where he actively dismantles his DEI teams, and where he smokescreens safeguards, my case demonstrates something inarguable: toxic and discriminatory environments aren’t just wrong, they’re anti-innovation. Hating women hurts everyone. The truth doesn’t matter to Mark Zuckerberg, but it matters to me.

For 15 years, I was trusted to lead Meta's most important teams and projects. I didn’t expect my career to end when I took a stand against a racist video game, but it’s true. When I was handed the opportunity to lead the company through the expansion of Horizon, I was thrilled. However, I was horrified to walk into a room of men wallpapering over rampant racism, a product riddled with performance issues, and violations of public policy. Worst of all, the victims were predominantly children. When I raised the problem, I became the problem – a pattern that silences women everywhere, every day. I’ve been sexually assaulted by a boss on a business trip. I’ve been denied promotions because acknowledging my success meant acknowledging men’s failures. I’ve been told to act "less smart," I’ve been retaliated against for doing my job. Some of you are shocked because you had no idea these things happened – some of you are shocked because you had no idea we could be honest about it. My privilege means I have the responsibility to speak up, but the discrimination and hostile work environment I experienced happens at every level, in every industry. It’s exponentially worse for women of color. It creates bad business outcomes that disproportionately harm those we should be most eager to protect, it widens the wealth gap, it puts lives at risk, and it’s against the law. It cost me my career, it almost cost my life. I’d been crotch grabbed, screamed at, told to have sex with my boss for a promotion. I survived all of it. Nothing broke me like a job where I had to tell powerful men "no." When tech companies push marginalized leaders out, they build dangerous products. This isn’t just an ethical problem. It’s a governance failure. It puts employees, shareholders, and users everywhere - especially those most in need of protection - at risk. When Meta's leadership dismissed and excluded me, they weren't just sidelining women, they were prioritizing power over people. They were putting their growth before the good. This cycle repeats itself across the industry: Women, minorities, neurodivergent people, raise ethical red flags about product safety They face retaliation and exclusion for speaking truth to power Harm then falls disproportionately on the most vulnerable users In the near-term, tech companies must strengthen protections for ethical whistleblowers, make promotion practices transparent, recommit to diversity programs and attach their success to executive compensation, and hardwire compliance into the product development process. Where Mark Zuckerberg rants about companies needing to be more masculine, where he actively dismantles his DEI teams, and where he smokescreens safeguards, my case demonstrates something inarguable: toxic and discriminatory environments aren’t just wrong, they’re anti-innovation. Hating women hurts everyone. The truth doesn’t matter to Mark Zuckerberg, but it matters to me.

Steve Soleimani

Alphabet's CFO Bets Big on Waymo as the Future of Transportation Alphabet CFO Anat Ashkenazi highlights Waymo's phenomenal autonomous vehicles and the company's strategic investments to scale up rapidly. Waymo is expanding to more markets, including Austin (via Uber), with Atlanta and Miami next. With over 200,000 paid robotaxi rides per week and a recent 5ドル.6 billion funding round, Waymo is valued at over 45ドル billion and is developing a sixth-generation driver to lower hardware costs. While consumer adoption is still growing, #Waymo is seen as the "clear technical leader" in the autonomous vehicle industry. https://lnkd.in/eVkYdzmN #GoogleCloud #Google #Alphabet #Cloud

Alphabet's CFO Bets Big on Waymo as the Future of Transportation Alphabet CFO Anat Ashkenazi highlights Waymo's phenomenal autonomous vehicles and the company's strategic investments to scale up rapidly. Waymo is expanding to more markets, including Austin (via Uber), with Atlanta and Miami next. With over 200,000 paid robotaxi rides per week and a recent 5ドル.6 billion funding round, Waymo is valued at over 45ドル billion and is developing a sixth-generation driver to lower hardware costs. While consumer adoption is still growing, #Waymo is seen as the "clear technical leader" in the autonomous vehicle industry. https://lnkd.in/eVkYdzmN #GoogleCloud #Google #Alphabet #Cloud

Asaf Israelit

🎉📣 Today’s a big day: Parter is officially out of stealth, and we’ve raised 5ドル.5M to reshape the future of hardware manufacturing. Since I started working in this industry, I kept running into the same problem: The most brilliant minds in hardware are stuck doing boring, repetitive tasks instead of building the future. Endless datasheets, spreadsheets, chasing lifecycles, compliance, tariffs, pricing shifts - you name it. What should be a strategic advantage turns into a daily fire drill. Together with Omer Gilat and Ronen Hoffer, we started Parter to change that, with a clear mission: To drive the future of hardware manufacturing by making it faster, more agile, and more creative. We built an AI platform for hardware teams - R&D, Engineering, and Supply Chain - that eliminates slow, manual, repetitive work. It helps teams make critical decisions, backed by real-time data and AI insights. So they can move faster, avoid risks, and stay focused on what only humans can do: imagine, design, and build. Huge thanks to our investors, our team, and the amazing customers already on this journey with us. We’re just getting started. 🚀🚀 Tal Slobodkin Esteban Reyes Shmil Levy Ariel Maislos Guy Schory Danielle Yanai Efi Cohen Ran Sarig Kfir Tishbi Raanan Raz Eyal Dagan Ofer Iny Nitzan Shapira Ran Ribenzaft Oren Buskila Evan Cummack Aviad Ben-laish Yuval Segev Elad Star Raz Zeevy Colin Campbell StageOne Ventures Zenda Capital Axios

🎉📣 Today’s a big day: Parter is officially out of stealth, and we’ve raised 5ドル.5M to reshape the future of hardware manufacturing. Since I started working in this industry, I kept running into the same problem: The most brilliant minds in hardware are stuck doing boring, repetitive tasks instead of building the future. Endless datasheets, spreadsheets, chasing lifecycles, compliance, tariffs, pricing shifts - you name it. What should be a strategic advantage turns into a daily fire drill. Together with Omer Gilat and Ronen Hoffer, we started Parter to change that, with a clear mission: To drive the future of hardware manufacturing by making it faster, more agile, and more creative. We built an AI platform for hardware teams - R&D, Engineering, and Supply Chain - that eliminates slow, manual, repetitive work. It helps teams make critical decisions, backed by real-time data and AI insights. So they can move faster, avoid risks, and stay focused on what only humans can do: imagine, design, and build. Huge thanks to our investors, our team, and the amazing customers already on this journey with us. We’re just getting started. 🚀🚀 Tal Slobodkin Esteban Reyes Shmil Levy Ariel Maislos Guy Schory Danielle Yanai Efi Cohen Ran Sarig Kfir Tishbi Raanan Raz Eyal Dagan Ofer Iny Nitzan Shapira Ran Ribenzaft Oren Buskila Evan Cummack Aviad Ben-laish Yuval Segev Elad Star Raz Zeevy Colin Campbell StageOne Ventures Zenda Capital Axios

David Noël-Romas

I recently left Stripe to co-found a new VC-funded startup—tada! I’ll share more soon, but right now I want to talk about the future of my beloved craft: computer programming. Because I know exactly who will thrive in this brave new world of generative coding. The latest trend for tech CEOs—of companies both large[0] and small[1]—is to loudly pronounce that they don’t expect to be hiring programmers for much longer. Is this an accurate projection of the future, or is it blustery bandwagon PR? These are the same folks who trotted out the Jevons paradox[2] when DeepSeek sent the NASDAQ into a nosedive. In a market like software—where demand can surge when costs drop—we shouldn’t expect efficiency gains to kill off coding jobs. That said, everything is changing—fast. I’m currently the ~only programmer at my new startup, but over the past few days, I’ve "written" ~40,000 lines of code...with 99% of it generated by AI. Does that mean I’ll never hire another programmer? Of course not. (In fact, if you’re a cracked engineer looking for your next gig, DM me ;) ) What I will loudly announce is that the economics of software are shifting—radically, inevitably, and permanently. LLMs can turn a 0x engineer into a 0.5x, and a 10x engineer into a 100x—or even 1,000x. It’s akin to jumping from assembly to high-level languages, but on a far larger scale: not only are existing programmers supercharged, but everyone with natural language skills gets a baseline ability to code. What does that mean? Expect highly personalized software at the 0.5x end—people building quick, custom solutions for themselves—and watch 100-person tech companies compete with (and displace) today’s 100,000-person giants at the 1,000x end. The realm of possibility has just expanded in every direction. The programmers who’ll thrive in this new world are the same ones who always do: those who delight in creating something from nothing, armed with just a keyboard. For them, the future is brighter than ever. Yes, this AI era could unseat big incumbents. But no, it won’t leave humans sidelined. Instead, we get to build a new age of prosperity—together. [0] https://lnkd.in/gEfrgJHk [1] https://lnkd.in/gbTzPatA [2] https://lnkd.in/gQ2KDDhi

I recently left Stripe to co-found a new VC-funded startup—tada! I’ll share more soon, but right now I want to talk about the future of my beloved craft: computer programming. Because I know exactly who will thrive in this brave new world of generative coding. The latest trend for tech CEOs—of companies both large[0] and small[1]—is to loudly pronounce that they don’t expect to be hiring programmers for much longer. Is this an accurate projection of the future, or is it blustery bandwagon PR? These are the same folks who trotted out the Jevons paradox[2] when DeepSeek sent the NASDAQ into a nosedive. In a market like software—where demand can surge when costs drop—we shouldn’t expect efficiency gains to kill off coding jobs. That said, everything is changing—fast. I’m currently the ~only programmer at my new startup, but over the past few days, I’ve "written" ~40,000 lines of code...with 99% of it generated by AI. Does that mean I’ll never hire another programmer? Of course not. (In fact, if you’re a cracked engineer looking for your next gig, DM me ;) ) What I will loudly announce is that the economics of software are shifting—radically, inevitably, and permanently. LLMs can turn a 0x engineer into a 0.5x, and a 10x engineer into a 100x—or even 1,000x. It’s akin to jumping from assembly to high-level languages, but on a far larger scale: not only are existing programmers supercharged, but everyone with natural language skills gets a baseline ability to code. What does that mean? Expect highly personalized software at the 0.5x end—people building quick, custom solutions for themselves—and watch 100-person tech companies compete with (and displace) today’s 100,000-person giants at the 1,000x end. The realm of possibility has just expanded in every direction. The programmers who’ll thrive in this new world are the same ones who always do: those who delight in creating something from nothing, armed with just a keyboard. For them, the future is brighter than ever. Yes, this AI era could unseat big incumbents. But no, it won’t leave humans sidelined. Instead, we get to build a new age of prosperity—together. [0] https://lnkd.in/gEfrgJHk [1] https://lnkd.in/gbTzPatA [2] https://lnkd.in/gQ2KDDhi

Supreet Deshpande

🚨The Real Story Behind "100% on USMLE"🚨 This weekend, our engineering team at SynthioLabs got curious. OpenEvidence announced they were the first AI to score 100% on the USMLE — big headline So we asked ourselves: could we do the same? 50+ runs. 1000ドル+ in compute and a few coffees later — we finally got that one run that hit 100%. ✅ But that led to the obvious question: how did OpenEvidence do it? One of our engineers said, "I bet they can’t reproduce the 100% themselves" So we put it to the test. We went to OpenEvidence, asked it one of the very same USMLE questions... ❌ And it got it wrong; and that’s the real story. In healthcare, you don’t get to claim perfection when lives are at stake. Because here’s the risk: physicians see "100% on USMLE" and start trusting AI blindly for diagnosis. That’s dangerous. 👉 (Link in comments: proof with the tested question + wrong OE output) At SynthioLabs, our thesis is clear: AI for Life Sciences isn’t about acing exams — it’s about trust. 🔒 Safety & compliance — every response grounded in credible, validated, regulatory-approved sources 🩺 Medical Info — supporting healthcare stakeholders with on-demand, evidence-based answers when it matters most ⚠️ No diagnosis — because in medicine, responsibility matters more than buzz. 100% on USMLE is a parlor trick. The real work is building AI that’s safe, credible, and usable in practice. That’s the standard we hold ourselves to. 🙏 We greatly respect OpenEvidence and the mission they’re on. At the same time, it’s important to clarify these discrepancies. Would welcome a conversation with their team.

🚨The Real Story Behind "100% on USMLE"🚨 This weekend, our engineering team at SynthioLabs got curious. OpenEvidence announced they were the first AI to score 100% on the USMLE — big headline So we asked ourselves: could we do the same? 50+ runs. 1000ドル+ in compute and a few coffees later — we finally got that one run that hit 100%. ✅ But that led to the obvious question: how did OpenEvidence do it? One of our engineers said, "I bet they can’t reproduce the 100% themselves" So we put it to the test. We went to OpenEvidence, asked it one of the very same USMLE questions... ❌ And it got it wrong; and that’s the real story. In healthcare, you don’t get to claim perfection when lives are at stake. Because here’s the risk: physicians see "100% on USMLE" and start trusting AI blindly for diagnosis. That’s dangerous. 👉 (Link in comments: proof with the tested question + wrong OE output) At SynthioLabs, our thesis is clear: AI for Life Sciences isn’t about acing exams — it’s about trust. 🔒 Safety & compliance — every response grounded in credible, validated, regulatory-approved sources 🩺 Medical Info — supporting healthcare stakeholders with on-demand, evidence-based answers when it matters most ⚠️ No diagnosis — because in medicine, responsibility matters more than buzz. 100% on USMLE is a parlor trick. The real work is building AI that’s safe, credible, and usable in practice. That’s the standard we hold ourselves to. 🙏 We greatly respect OpenEvidence and the mission they’re on. At the same time, it’s important to clarify these discrepancies. Would welcome a conversation with their team.

Ameer Haj Ali, PhD

Unpopular opinion: The RL hype is a red flag for technical depth. When I hear founders or VCs casually dropping "reinforcement learning" in pitches, it's usually code for "we don't actually understand our problem space." Here's the reality check: RL only works when you have: - Perfect simulation environments - Extremely fast iteration cycles - Clear reward functions - Tolerance for massive compute costs That's why it succeeds in games (perfect simulation) and search optimization (Massive compute capacity at Deepmind AlphaGo/Anthropic/OpenAI) but fails spectacularly in the messy, slow-feedback real world. The graveyard of "RL-powered" startups tells the story. They burned through $$$$ trying to apply algorithms to problems that needed simple heuristics or supervised learning. Most of the time, a well-tuned evolutionary algorithm will outperform your fancy RL setup while using <1/10th the compute and delivering results you can actually explain to stakeholders. The next time someone pitches you an RL solution, ask them: "Have you tried solving this with a basic optimization approach first?" Watch how quickly the conversation changes. I am surprised when people are surprised that Direct Preference Optimization (DPO) can exceed PPO-based RLHF [1] or Reflective Prompt Evolution Can Outperform Reinforcement Learning [2]. [1] https://lnkd.in/gCY3bcTQ [2] https://lnkd.in/gQRwyQyM

Unpopular opinion: The RL hype is a red flag for technical depth. When I hear founders or VCs casually dropping "reinforcement learning" in pitches, it's usually code for "we don't actually understand our problem space." Here's the reality check: RL only works when you have: - Perfect simulation environments - Extremely fast iteration cycles - Clear reward functions - Tolerance for massive compute costs That's why it succeeds in games (perfect simulation) and search optimization (Massive compute capacity at Deepmind AlphaGo/Anthropic/OpenAI) but fails spectacularly in the messy, slow-feedback real world. The graveyard of "RL-powered" startups tells the story. They burned through $$$$ trying to apply algorithms to problems that needed simple heuristics or supervised learning. Most of the time, a well-tuned evolutionary algorithm will outperform your fancy RL setup while using <1/10th the compute and delivering results you can actually explain to stakeholders. The next time someone pitches you an RL solution, ask them: "Have you tried solving this with a basic optimization approach first?" Watch how quickly the conversation changes. I am surprised when people are surprised that Direct Preference Optimization (DPO) can exceed PPO-based RLHF [1] or Reflective Prompt Evolution Can Outperform Reinforcement Learning [2]. [1] https://lnkd.in/gCY3bcTQ [2] https://lnkd.in/gQRwyQyM

Abhishek Agarwal

Waymo's CEO just shared fascinating insights about the state of autonomous vehicles in 2025: The company delivered 4 million paid rides in 2024 alone - more than all previous years combined. They're now doing 150,000 rides weekly and driving 1 million miles per week (more than a human drives in a lifetime). 2024 was their breakthrough year. They launched the first fully autonomous airport service at Phoenix Sky Harbor - anyone can now land and immediately hail a driverless car through the Waymo app. Unlike Tesla which relies solely on cameras, Waymo uses a multi-sensor approach combining lidar, radar, and cameras. They argue this redundancy is essential for safety - and they're the only company offering 24/7 service for the past 4 years. Their expansion plans are aggressive: They're partnering with Uber to launch in Austin and Atlanta They're testing in Miami with plans to launch there They're entering Japan, starting with Tokyo They've announced a partnership with Hyundai for 2026-2027 The safety data is compelling: Compared to human drivers, Waymo vehicles had 78% fewer injury-causing crashes and 81% fewer airbag deployments across 33 million miles driven. A Swiss Re study showed even more dramatic results: 92% reduction in personal injury claims and 80% reduction in property damage claims. The future isn't about perfect autonomous vehicles - it's about being significantly safer than human drivers. And Waymo's data suggests we're already there. This isn't a winner-takes-all market. Waymo sees a future where their technology powers ride-hailing, local delivery, personal vehicles, and long-haul trucking through partnerships rather than doing it all themselves. They're proving that autonomous driving isn't science fiction anymore - it's a safer, more efficient reality that's already here. https://lnkd.in/gM54MGVD

Waymo's CEO just shared fascinating insights about the state of autonomous vehicles in 2025: The company delivered 4 million paid rides in 2024 alone - more than all previous years combined. They're now doing 150,000 rides weekly and driving 1 million miles per week (more than a human drives in a lifetime). 2024 was their breakthrough year. They launched the first fully autonomous airport service at Phoenix Sky Harbor - anyone can now land and immediately hail a driverless car through the Waymo app. Unlike Tesla which relies solely on cameras, Waymo uses a multi-sensor approach combining lidar, radar, and cameras. They argue this redundancy is essential for safety - and they're the only company offering 24/7 service for the past 4 years. Their expansion plans are aggressive: They're partnering with Uber to launch in Austin and Atlanta They're testing in Miami with plans to launch there They're entering Japan, starting with Tokyo They've announced a partnership with Hyundai for 2026-2027 The safety data is compelling: Compared to human drivers, Waymo vehicles had 78% fewer injury-causing crashes and 81% fewer airbag deployments across 33 million miles driven. A Swiss Re study showed even more dramatic results: 92% reduction in personal injury claims and 80% reduction in property damage claims. The future isn't about perfect autonomous vehicles - it's about being significantly safer than human drivers. And Waymo's data suggests we're already there. This isn't a winner-takes-all market. Waymo sees a future where their technology powers ride-hailing, local delivery, personal vehicles, and long-haul trucking through partnerships rather than doing it all themselves. They're proving that autonomous driving isn't science fiction anymore - it's a safer, more efficient reality that's already here. https://lnkd.in/gM54MGVD

Jonathan Chang

Observations in Silicon Valley Preseed rounds done in SF has a 2-4x markup versus deals done outside of SF Met 2 companies building the exact same product & with similar rev < 100ドルk Company A: SF based, younger founders out of college with less experience in the space, TS at 25ドルm post Company B: Midwest-based, older founders with connections in the industry, TS at 6ドルm post Same product, same traction...that’s the SF premium

Paul Jellison

I talk about this with friends and many tell me I'm wrong, mostly on the timescale. Autonomous Vehicles are scaling now. Waymo is out ahead, but Tesla certainly has the capability and a strategy. With Waymo and Tesla going for the robotaxi market in direct competition, I think we can expect every populated area to have robotaxi options by 2030. With scale comes economies of scale. For ride hailing with a human, cost per mile is 2ドル+. For a personal vehicle, cost per mile is 70 cents. Elon says they have a path to 40 cents. As costs compress, many people will want to remove the hassle of driving from their lives. Families might move to one car, many will choose to rely strictly on AVs. By 2035, it's clear to everyone that AVs are both much safer and much cheaper than owning, maintaining, and manually driving their car. Public safety will become the primary driver for adoption, and governments will likely fund and incentivize the final transition. At this point, there are no new people seeking out or even wanting to get a driver's license. Driving has become more of a hobby for enthusiasts than a necessity for daily life. And with cars no longer needing dedicated storage, home owners will repatriate their garages as living space. I experienced it first hand this week. In 10 years, how we live will be totally transformed by this tech.

I talk about this with friends and many tell me I'm wrong, mostly on the timescale. Autonomous Vehicles are scaling now. Waymo is out ahead, but Tesla certainly has the capability and a strategy. With Waymo and Tesla going for the robotaxi market in direct competition, I think we can expect every populated area to have robotaxi options by 2030. With scale comes economies of scale. For ride hailing with a human, cost per mile is 2ドル+. For a personal vehicle, cost per mile is 70 cents. Elon says they have a path to 40 cents. As costs compress, many people will want to remove the hassle of driving from their lives. Families might move to one car, many will choose to rely strictly on AVs. By 2035, it's clear to everyone that AVs are both much safer and much cheaper than owning, maintaining, and manually driving their car. Public safety will become the primary driver for adoption, and governments will likely fund and incentivize the final transition. At this point, there are no new people seeking out or even wanting to get a driver's license. Driving has become more of a hobby for enthusiasts than a necessity for daily life. And with cars no longer needing dedicated storage, home owners will repatriate their garages as living space. I experienced it first hand this week. In 10 years, how we live will be totally transformed by this tech.

Sam Bhagwat

Mastra has a new agent orchestration layer! Mastra's always used Vercel's AI SDK for model routing and streaming. With the recent release of AI SDK v5, rather than shipping a breaking change, we decided to upgrade our agent orchestration capabilities and add a streaming layer so our users didn't have to upgrade. This implementation lets us take on more responsibility for the agent-specific parts. Some highlights: - We have increased control over the tool calling and agent loop - We now handle both v4 and v5 message formats simultaneously. The playground model switcher auto-detects versions. - Our new streamVNext and generateVNext APIs are able to output v5 streams. Blog post in comments

Mastra has a new agent orchestration layer! Mastra's always used Vercel's AI SDK for model routing and streaming. With the recent release of AI SDK v5, rather than shipping a breaking change, we decided to upgrade our agent orchestration capabilities and add a streaming layer so our users didn't have to upgrade. This implementation lets us take on more responsibility for the agent-specific parts. Some highlights: - We have increased control over the tool calling and agent loop - We now handle both v4 and v5 message formats simultaneously. The playground model switcher auto-detects versions. - Our new streamVNext and generateVNext APIs are able to output v5 streams. Blog post in comments