Lenny RAG MCP Server

{ "episode": { "guest": "Eric Simons", "expertise_tags": [ "AI code generation", "Product development", "Startup scaling", "WebAssembly", "Browser-based development", "Founder/CEO", "Text-to-app tools" ], "summary": "Eric Simons, co-founder and CEO of StackBlitz, discusses the explosive growth of Bolt, an AI-powered text-to-app builder that reached 40 million ARR in just five months after launch. The episode explores how seven years of building WebContainer technology (an operating system that runs in the browser) enabled Bolt's success. Eric shares insights on scaling with a small 15-20 person team, the importance of founding team stability, why Claude Sonnet was the inflection point for AI code quality, and how product managers and non-technical founders are now best positioned to thrive in an AI-first development world. He discusses upcoming features like Figma-to-code integration and a Slack bot developer, and reflects on the broader organizational and skill shifts coming to software development.", "key_frameworks": [ "Technology-first strategy with problem-finding later", "Staying alive longer = more shots on goal", "Team stability over headcount growth", "Low burn rate enables long-term bets", "Context per head over team size", "High-trust culture with agency", "Building in deterministic domains for AI advantage", "PM/designer-led product development future" ] }, "topics": [ { "id": "topic_1", "title": "Bolt's Explosive Growth: Zero to 40M ARR in 5 Months", "summary": "Eric reveals staggering growth numbers: Bolt launched with a tweet and went from zero to 20 million ARR in two months, crossed 30 million, and is approaching 40 million. By month four, they had 3 million registered users and 1 million monthly active users. This represents growth rates never seen in startup history, comparable only to Cursor.", "timestamp_start": "00:00:00", "timestamp_end": "00:08:33", "line_start": 1, "line_end": 87 }, { "id": "topic_2", "title": "Bolt Product Demo: Capabilities and Competitive Advantages", "summary": "Eric demonstrates Bolt's core functionality: a simple text box interface for building full-stack web and mobile apps from prompts. Key differentiator is speed and reliability powered by WebContainer, an OS running in the browser using CPU. Shows Spotify clone creation in 60 seconds, mobile app support via Expo integration with QR code preview, and native app deployment without technical knowledge.", "timestamp_start": "00:04:54", "timestamp_end": "00:18:59", "line_start": 55, "line_end": 185 }, { "id": "topic_3", "title": "Seven-Year Journey: WebContainer Technology Foundation", "summary": "Eric explains the seven-year technology bet that made Bolt possible. In 2016-2017, after being inspired by Figma's browser-based approach, he and co-founder Albert decided to build a WebAssembly-based operating system for web development. This required years of foundational work to make reliable and fast enough to support AI code generation at scale.", "timestamp_start": "00:19:09", "timestamp_end": "00:26:43", "line_start": 187, "line_end": 228 }, { "id": "topic_4", "title": "Founder Philosophy: Conservative Capital, Long Bets, and Survival", "summary": "Eric discusses how StackBlitz bootstrapped for 2-3 years and maintained low burn rate even after fundraising. They resisted pressure to hire aggressively during 2020-2021 when growth theater was fashionable. This allowed them to take long-term bets on technology and be alive when the opportunity (Bolt) emerged. The core lesson: maintain low burn, take multiple shots on goal, and survive long enough to find what works.", "timestamp_start": "00:25:04", "timestamp_end": "00:29:37", "line_start": 213, "line_end": 245 }, { "id": "topic_5", "title": "Launch Day and Product-Market Fit Signals", "summary": "Bolt launched with a single tweet and hit 60K ARR on day one (10% of company's prior annual revenue). Growth continued day two with 80K ARR. The team was unprepared for scale: servers melting, Anthropic ran out of GPUs, no mobile responsive view despite rapid user growth, and pricing needed complete overhaul as users burned through $9/month plans in 48 hours. The 90-day MVP had many missing basics.", "timestamp_start": "00:29:35", "timestamp_end": "00:34:08", "line_start": 241, "line_end": 281 }, { "id": "topic_6", "title": "Small Team, Massive Scale: How 15-20 People Managed Growth", "summary": "The secret to scaling with minimal headcount: 7-year founding team stability (rare in Silicon Valley), high trust, context per person instead of team size, agency to act independently across full stack (engineering to customer support), no political approval layers, and strong interpersonal bonds. Engineers moved from customer calls to fixing issues to landing UI changes in real-time without coordination.", "timestamp_start": "00:34:08", "timestamp_end": "00:39:28", "line_start": 280, "line_end": 305 }, { "id": "topic_7", "title": "Daily All-Hands Sync: Communication During Hypergrowth", "summary": "The entire company (15-20 people) meets on Zoom every day at 8 AM Pacific for at least an hour. They review everything front-to-back with zero fidelity loss in communication. This practice became critical after Bolt launched and continued through hypergrowth. While not sustainable forever, it ensures everyone knows what's happening and makes better independent decisions. Higher communication cost than typical, but zero information loss.", "timestamp_start": "00:41:28", "timestamp_end": "00:46:45", "line_start": 328, "line_end": 387 }, { "id": "topic_8", "title": "Prioritization Strategy: Balancing Triage vs. Big Bets", "summary": "Eric uses gut instinct for prioritization, balancing reactive bug fixes with proactive bets on features nobody is explicitly asking for. Example: mobile app support seemed controversial internally but became their biggest launch. Uses restaurant chef analogy: balance customer feedback ('this didn't taste good') with creator intuition ('I cooked something interesting'). This ability builds through years of getting kicked by the free market.", "timestamp_start": "00:41:01", "timestamp_end": "00:44:02", "line_start": 325, "line_end": 346 }, { "id": "topic_9", "title": "PRDs and Documentation: Keeping Specs Light", "summary": "StackBlitz uses Notion for PRDs but keeps them minimal with just enough context to align the team on key outcomes. Lengthy PRDs get glossed over. Prefers linking to working Bolt prototypes: 'a live actual demo is worth millions.' This shift reflects that building high-fidelity prototypes in Bolt is now faster and cheaper than static Figma mockups, changing how product development actually works.", "timestamp_start": "00:46:45", "timestamp_end": "00:49:08", "line_start": 383, "line_end": 401 }, { "id": "topic_10", "title": "Real-World Use Cases: From Prototype to Production", "summary": "Paul, a non-technical entrepreneur, built a production CRM with AI and Stripe billing in 3 weeks using Bolt, vs. 6 months and $30K from an agency. He spent $300 and is selling the product. Companies building greenfield projects can create production-grade software in days/weeks instead of months. Existing companies use Bolt for marketing sites, admin panels, and product acceleration. Integration with existing large codebases (1000+ files) not yet reliable.", "timestamp_start": "00:49:08", "timestamp_end": "00:52:39", "line_start": 401, "line_end": 430 }, { "id": "topic_11", "title": "Limitations of Bolt Today and Current Scope", "summary": "Main limitation: large existing codebases (1000+ files) won't work reliably yet due to LLM context limits. Users need developers for that (Cursor is better). Outside of that, requires learning curve like any tool (Photoshop, Figma). Most successful non-developer users are strong PMs who understand how to spec things clearly and debug issues. Engineering skills help but aren't required.", "timestamp_start": "00:52:24", "timestamp_end": "00:54:08", "line_start": 427, "line_end": 435 }, { "id": "topic_12", "title": "The Future of PMs: From Specs to Builders", "summary": "67% of Bolt users are non-developers (PMs, designers, entrepreneurs). This is reshaping the entire software world. PMs traditionally spec work for developers; now they'll directly code (prompt) to build. Developers will move to intellectually challenging tasks and debugging, no longer on cookie-cutter UI work. Org charts will change: fewer frontend engineers, more product/design leads with 1-2 engineers per pod reviewing code.", "timestamp_start": "00:54:54", "timestamp_end": "01:00:19", "line_start": 436, "line_end": 478 }, { "id": "topic_13", "title": "Skills for the Future: What to Learn and Avoid", "summary": "Don't learn to code just for job security; instead understand how systems work conceptually. Follow intrinsic interests (compilers, systems, etc.) if genuinely interested. The future won't require college degrees for developers—explore free online resources first. Learn problem-solving, articulation, and growth—these skills matter more. Eric and his co-founder didn't attend college and bootstrapped successfully.", "timestamp_start": "01:00:19", "timestamp_end": "01:05:13", "line_start": 476, "line_end": 520 }, { "id": "topic_14", "title": "Bolt Builders: Human Support for AI-Generated Code", "summary": "Announced a new program where non-developer users can connect with certified Bolt Builders experts (like Apple's Genius Bar) when the AI gets stuck. Users pay ~$50/hour for live chat support. This role will evolve as models improve, but currently offers a valuable service for unblocking users. Over time, models will get better, reducing need for this service, but it represents a human-in-the-loop approach to AI development.", "timestamp_start": "01:05:03", "timestamp_end": "01:06:58", "line_start": 517, "line_end": 532 }, { "id": "topic_15", "title": "Sonnet as the Inflection Point: Why Code Generation Works Now", "summary": "Claude Sonnet (launched June 2024) was the critical breakthrough model that made production-grade code generation possible. Bolt was attempted a year earlier with frontier models but failed—code output was unreliable. When they got a peek at Sonnet in May 2025, they revived the project. This flip of AI-first (then human reviewer) vs. human-first (then AI helper) only became possible with Sonnet's coding ability.", "timestamp_start": "01:06:50", "timestamp_end": "01:12:05", "line_start": 530, "line_end": 564 }, { "id": "topic_16", "title": "Why LLMs Will Get Better at Code Than Any Other Domain", "summary": "Software is deterministic: code either runs or it doesn't, unlike law, medicine, or other domains where judgment and context vary. This allows massive permutation of training data. Anthropic's insight: deep reinforcement learning on millions of app variations per second builds best-in-world code models. Tech companies (Google, OpenAI, etc.) are now gunning for coding since the market opportunity is trillions of dollars—the world runs on software.", "timestamp_start": "01:08:11", "timestamp_end": "01:12:40", "line_start": 538, "line_end": 573 }, { "id": "topic_17", "title": "Upcoming Features: Figma-to-Code and Slack Bot Integration", "summary": "Two major features launching soon: (1) Figma integration: put bolt.new/ in front of any Figma URL to convert designs to full-stack apps, powered by Anima. (2) Slack bot: invite @Bolt to thread, ask it to build things by referencing Figma designs and requirements in conversation history. Bot automatically converts designs and implements changes. Moves product development into Slack.", "timestamp_start": "01:14:33", "timestamp_end": "01:18:40", "line_start": 611, "line_end": 665 }, { "id": "topic_18", "title": "How Bolt's Team Builds Bolt: Using Cursor, Claude, and Bolt", "summary": "StackBlitz engineers use Cursor as primary AI coding tool for editing large existing codebases. They heavily use Bolt itself for product development and prototyping. When designs need to be pixel-perfect, team goes to Figma, then pulls designs into Bolt for implementation. They use subscriptions to Claude and ChatGPT but consider Cursor the main development tool. Few AI tools end up being actually useful in practice.", "timestamp_start": "01:18:59", "timestamp_end": "01:20:14", "line_start": 664, "line_end": 680 }, { "id": "topic_19", "title": "Advice for New Bolt Users: Treat It Like a Developer Coworker", "summary": "Talk to Bolt like you'd write a Linear/JIRA ticket for a developer: be specific on what matters, let it be creative on vibes. For first-time users without an idea, build a personal website from your LinkedIn bio—this zero-shot experience (beautiful site in minutes) demonstrates the power. Taking time to craft clear requirements upfront prevents getting stuck with a bad first iteration.", "timestamp_start": "01:20:14", "timestamp_end": "01:22:55", "line_start": 682, "line_end": 705 }, { "id": "topic_20", "title": "From Homeless to Hypergrowth: The AOL Office Squat Story", "summary": "In 2012 at age 19, Eric lived in the AOL office in Palo Alto for 4-5 months while building a K12 education startup through Imagine K12. After running out of money (YC only gave $20K), he realized the office had couches, food, gym, shower, and laundry. Security guards befriended him thinking he was working hard. Eventually got kicked out at 4 AM, slept outside a Starbucks, lived on $1/day (McDonald's Dollar Menu), contacted a fellow entrepreneur for a couch. Press covered it as a worldwide story.", "timestamp_start": "01:23:18", "timestamp_end": "01:27:01", "line_start": 707, "line_end": 740 } ], "insights": [ { "id": "i1", "text": "The rate you're growing is absurd. You're in this cohort of companies that are just growing at rates that we've never seen in the history of startups.", "context": "Lenny opening observation about Bolt's unprecedented growth trajectory", "topic_id": "topic_1", "line_start": 1, "line_end": 2 }, { "id": "i2", "text": "Most importantly, it's been the people. It's rare to find startups where you have the core group of five, six, seven people that have been there for five years plus.", "context": "Eric explaining why a 15-20 person team can manage million-user scale growth", "topic_id": "topic_6", "line_start": 11, "line_end": 11 }, { "id": "i3", "text": "You basically were building a tech first, and then looking for a problem to solve later, which is often what people tell you not to do. There are periods of time where you have to make judgment calls that are not going to be the consensus view. You got to have confidence in your convictions on how to best play the hand.", "context": "Addressing the counter-intuitive strategy of tech-first product development", "topic_id": "topic_4", "line_start": 14, "line_end": 17 }, { "id": "i4", "text": "Figma was also a browser-based deep technology play. Their first pitch for Figma, they didn't have a design tool. Their first pitch was this 3D ball dropping into water, inside of a browser. The pitch basically was, 'Browsers have this new capability called WebGL, the predecessor to WebAssembly, and with these things, for the first time, you could actually create a graphics rendering engine that you could then build a design tool on top of.'", "context": "Eric explaining how Figma's success inspired WebContainer bet", "topic_id": "topic_3", "line_start": 194, "line_end": 197 }, { "id": "i5", "text": "It's also, if you look at the other productivity apps that have really worked on the web, they've all had this compute model, right? Figma, when you open a Figma document, there's not like some cloud VM that gets spun up for you to render the documents. You're dragging things around. It's using your CPU and your memory to do the work. Same thing with Google Docs. That's the only model that's ever scaled to a billion users.", "context": "Explaining why local compute instead of cloud VMs is essential for billion-user scale", "topic_id": "topic_3", "line_start": 202, "line_end": 206 }, { "id": "i6", "text": "Unless you're seeing pull, just people pulling the product out of your hands, you don't want to be spending money. You should be like, default, no. And when you go and buy software, you should be going, 'We're a tiny startup. Can you sell it for half?' Everything you buy, just keep the burn rate as low as possible, because you need as many shots on goal as you can possibly get.", "context": "Core principle of StackBlitz's capital efficiency strategy", "topic_id": "topic_4", "line_start": 232, "line_end": 233 }, { "id": "i7", "text": "I think that's the hard thing about being an entrepreneur. There are periods of time where you have to make judgment calls that are not going to be the consensus view. Maybe years later, it'll become the consensus view, but you got to have confidence in your convictions on how to best play the hand.", "context": "Eric on the psychological challenge of contrarian founder decisions", "topic_id": "topic_4", "line_start": 239, "line_end": 239 }, { "id": "i8", "text": "Anthropic ran out of GPUs for us. Dario emailed me, he was like, 'Listen, we don't have anything more to give you.' It was just bananas, for weeks. It felt like in 300, when they're surrounded by 10,000 people, and our team is just doing everything.", "context": "Example of scale challenges: infrastructure constraints weren't the limiting factor", "topic_id": "topic_5", "line_start": 269, "line_end": 269 }, { "id": "i9", "text": "High trust, and people, we all just have enjoyed working together in the past. Maybe that's why, that's the only reason that anyone would ever stay at a company for that long, or whatever. And so those sorts of stressful situations, I think, are make or break. Those are make or break for any team.", "context": "Why long-tenured teams outperform during hypergrowth crises", "topic_id": "topic_6", "line_start": 293, "line_end": 293 }, { "id": "i10", "text": "People, of course it's people have a career trajectory, and that sort of thing, but they really are motivated by just working on cool things, and are chucking their ego at the door. And they're there to collectively build something great, not just kind of follow, and be the brilliant jerk.", "context": "Culture hiring principle: intrinsic motivation over titles or ego", "topic_id": "topic_6", "line_start": 298, "line_end": 298 }, { "id": "i11", "text": "The thing about just having everyone in the same room every day is that, a lot of people will complain that it's the most expensive use of everyone's time, but it's like, 'Yep. But there's 0% fidelity loss in that. Everything, every day, is being audited front to back, and being discussed front to back.'", "context": "Defense of daily all-hands during hypergrowth despite communication overhead", "topic_id": "topic_7", "line_start": 365, "line_end": 365 }, { "id": "i12", "text": "There's some amount of feedback from the customers of, 'This thing didn't taste good.' And then there's like, 'Hey, we've been cooking something interesting, and this tastes... I think you're going to like this. I think this is a killer dish.' And so you kind of have to balance those things.", "context": "Restaurant chef analogy for balancing reactive customer feedback with proactive vision", "topic_id": "topic_8", "line_start": 338, "line_end": 338 }, { "id": "i13", "text": "I think it's actually, largely, a function of just years of experience doing it. I think if you kind of rewound 10 years ago, I would have had really no, I wouldn't have had just the years of getting my butt kicked by the free market, to have cultivated a sense of this stuff.", "context": "Prioritization ability comes from accumulated market feedback over years", "topic_id": "topic_8", "line_start": 341, "line_end": 341 }, { "id": "i14", "text": "If a picture is worth a thousand words, a live actual demo is worth millions. You can feel it. It's real.", "context": "Why Bolt prototypes are now more valuable than static design mockups", "topic_id": "topic_9", "line_start": 398, "line_end": 398 }, { "id": "i15", "text": "Before, it was too expensive. The idea of, 'And let's prototype it, the engineers' code a proto...' It's like, that, it would take forever. It would be expensive. And now it's faster to do this with Bolt, in code, and have a real working software product, than dragging around frames and Figma to actually make a static version of it.", "context": "Cost/speed reversal: coding prototypes now cheaper than design mockups", "topic_id": "topic_9", "line_start": 398, "line_end": 398 }, { "id": "i16", "text": "You're going to spend a couple of days, weeks, whatever. But the cost reduction there, 30 grand versus $300. It's 99% cheaper. Six months versus three weeks. I mean, it's like order of magnitude sort of fashioned delivery on the thing.", "context": "Quantified ROI of Bolt for greenfield projects: 99% cost savings, 2-3x faster delivery", "topic_id": "topic_10", "line_start": 415, "line_end": 416 }, { "id": "i17", "text": "The people that we see that are most successful with Bolt, outside of developers, the people we see that are most successful are people that are amazing PMs. Because these are people that understand enough about how the technology works, typically, and their job is to direct developers on how to go and improve the product.", "context": "PMs have natural advantage in using AI code generation tools", "topic_id": "topic_11", "line_start": 434, "line_end": 434 }, { "id": "i18", "text": "When all these tools came out, there was so many people saying, 'Okay, PMs are dead. We don't need them anymore. We can just build things so quickly and easily, what's the point?' But the hard part now is, 'What the hell should we build? Can we clearly articulate what it is we want to build?' And then, 'Can we just have the taste to know, is this right, is this correct? Is this good? Is this going to solve the problem?'", "context": "PMs remain critical because the bottleneck shifted from building to deciding what to build", "topic_id": "topic_12", "line_start": 437, "line_end": 437 }, { "id": "i19", "text": "67% of our users are not developers, at this point. And when I started talking to these folks, it just kind of clicked as like, 'Oh, well, this is going to change everything. The entire software world order is going to get rewritten, here.'", "context": "Majority of Bolt users are non-technical, signaling fundamental shift in development", "topic_id": "topic_12", "line_start": 440, "line_end": 440 }, { "id": "i20", "text": "PMs, they're going to be 'writing code', quote, unquote, instead of just writing a JIRA ticket and waiting for a developer to do it. The developers are going to be able to work on intellectually challenging tasks that LLMs are not well suited for, and still being augmented by LLMs to do it. But PMs are going to be able to go in and just make the changes themselves.", "context": "Vision of PM-driven development where PMs prompt instead of spec", "topic_id": "topic_12", "line_start": 443, "line_end": 443 }, { "id": "i21", "text": "What blew my mind is, it's not priced in, to any of these companies out there. And it's not reflected in the org charts of all the software companies in the world right now. That is going to completely change.", "context": "Org structure changes from PM-developer split to PM-coder teams not yet reflected in market", "topic_id": "topic_12", "line_start": 446, "line_end": 446 }, { "id": "i22", "text": "Understanding how to leverage these AI tools is key. I wouldn't necessarily, I think maybe getting a basic understanding of how programming works, et cetera, is great. But it doesn't have to be, because I think back to if Bolt existed, like Albert and I say this to each other all the time. Since the get-go, StackBlitz, we've been building the thing that we wish we had when we were 13. I mean, I don't know if I would've gone as deep as I did on learning how to code, and being an engineer, if that had been around then.", "context": "Deep coding knowledge no longer a prerequisite if AI tools existed earlier", "topic_id": "topic_13", "line_start": 479, "line_end": 485 }, { "id": "i23", "text": "I think people need to follow their intrinsic interests. If folks are really interested in really getting in the nitty-gritty of how computers work, and program leaders, or spark and compilers, or whatever, go for it.", "context": "Pursue deep technical knowledge only if genuinely interested, not for job security", "topic_id": "topic_13", "line_start": 488, "line_end": 488 }, { "id": "i24", "text": "I think that stuff is still going to be relevant. I don't know if we're going to really have, we'll see, but to the degree that there's AGI where it's like, we don't have to think about anything ever again... I think from, at least what I feel like seems like the next at least five years of what we're looking at, I think people are still going to, there's still going to be places to specialize, and really go deep.", "context": "5-year outlook: specialization will matter, but unclear beyond that with AGI", "topic_id": "topic_13", "line_start": 488, "line_end": 494 }, { "id": "i25", "text": "You want to go into it with the idea, not like, 'I'm going to go and learn computer science because I'm going to get a job for sure out of it.' I just think that's generally not a good approach.", "context": "Don't pursue technical education purely for employment guarantees", "topic_id": "topic_13", "line_start": 494, "line_end": 494 }, { "id": "i26", "text": "It costs nothing to go and explore and learn for free, online.", "context": "Abundance of free learning resources makes expensive college debt risky", "topic_id": "topic_13", "line_start": 506, "line_end": 506 }, { "id": "i27", "text": "Get good at figuring out what people need and want, what problems they need solved. Get good at articulating it really well to the AI tools. You don't need to be a great prompt engineer, but being clear about it, it'll save you a lot of time.", "context": "Core skill for AI-enabled product development: problem definition and clear specification", "topic_id": "topic_14", "line_start": 509, "line_end": 512 }, { "id": "i28", "text": "We actually, two weeks ago I think, we announced this program called Bolt Builders. And it's basically the genius bar at the Apple Store, where as folks are building on Bolt that are not developers, they'll run into some nook or cranny where the AI just cannot figure it out.", "context": "New service model: human experts help unstick AI-generated code for non-developers", "topic_id": "topic_14", "line_start": 521, "line_end": 521 }, { "id": "i29", "text": "Engineers get to focus on difficult challenges, not like cookie cutter, 'Let's make another CRUD app,' stuff. They get to, debugging is challenging, and fun. And going and working on intellectually stimulating tasks.", "context": "Future of engineering: shifted from routine UI work to debugging and complex challenges", "topic_id": "topic_14", "line_start": 527, "line_end": 527 }, { "id": "i30", "text": "Now it's AI first, and person second. Versus person building the thing, and then AI. It's like a human Copilot is flipping it.", "context": "Paradigm flip: AI generates 99%, human reviews 1%, versus old model (human writes, AI suggests)", "topic_id": "topic_15", "line_start": 530, "line_end": 530 }, { "id": "i31", "text": "Sonnet was really the first model that flipped the equation, because that was really us, and old Cursor, and all these other things. The rapid growth started the second Sonnet went online. We actually tried building Bolt almost exactly a year ago, with the frontier models at the time. Spent a week or two building it. It just didn't work. The output, the code output was not reliable enough.", "context": "Sonnet as the specific inflection point that unlocked viable AI code generation at scale", "topic_id": "topic_15", "line_start": 533, "line_end": 533 }, { "id": "i32", "text": "What's going on here is, a very critical threshold has been passed with LLM's ability to write production grade code and apps that actually look beautiful, and actually function well. It's not perfect, but there's kind of this zero to one moment that's happened where it's like, 'Okay, so now, yeah. Now the AI is the first thing,' and then you're kind of popping a developer in every now and then, versus the other way around.", "context": "Describes the zero-to-one inflection point in AI code generation capability", "topic_id": "topic_15", "line_start": 536, "line_end": 536 }, { "id": "i33", "text": "Software is deterministic. When you write code and you hit run, it either runs or it doesn't. And that's the key insight Anthropic really had. They just went deep. And then, this is what they're doing, is just reinforcement learning on basically permutating every type of app you could ever build, and just spinning up tens of thousands of cores or whatever to do that.", "context": "Why software is the ideal domain for LLM improvement: deterministic outcomes enable infinite training data generation", "topic_id": "topic_16", "line_start": 560, "line_end": 560 }, { "id": "i34", "text": "Making their LLMs the best in the world at building beautiful, reliable applications. I'm extremely bullish. It makes technical sense why, of anything, LLMs are going to get insanely better at writing code than probably most other types of applications for LLMs. Simply because it's something that can be extremely deterministic, and permutated thousands and thousands and thousands of times per second.", "context": "Anthropic's technical approach explains why coding is LLM advantage over other domains", "topic_id": "topic_16", "line_start": 560, "line_end": 560 }, { "id": "i35", "text": "And so I think the broader trend here is... And Sonnet has woken everyone up. Google, and Open AI, all the... Everyone is now gunning for coding because, how big is the market opportunity to rewrite the software world order? It's trillions of dollars, or something, right? The world runs on software.", "context": "Market-wide competition emerging in coding AI as companies recognize the trillion-dollar opportunity", "topic_id": "topic_16", "line_start": 563, "line_end": 563 }, { "id": "i36", "text": "I did not know that. I didn't realize that so much of this was unlocked with, like it's sitting on top of Anthropic work, and specifically Sonnet. That was the first model, you're saying, that could code well enough.", "context": "Lenny recognizing that Bolt's existence is entirely contingent on Anthropic's Claude Sonnet model", "topic_id": "topic_16", "line_start": 551, "line_end": 551 }, { "id": "i37", "text": "When we did raise money, we barely spent it. Largely because it was like, 'We need to just take a lot of smart bets, and it doesn't make sense.' Until you see pull, just people pulling the product out of your hands, you don't want to be spending money.", "context": "Capital efficiency principle: raise money but maintain low burn to maximize option value", "topic_id": "topic_4", "line_start": 230, "line_end": 233 }, { "id": "i38", "text": "Talk to this thing like you do a Linear ticket, or a JIRA ticket. That would be my advice. And talk to this like you would, like you're talking to one of the developers on your team. And what that means is, be specific on things that matter. And on things where, also, you can let it be creative.", "context": "Best practice for prompting Bolt: balance specificity with creative freedom", "topic_id": "topic_19", "line_start": 692, "line_end": 692 }, { "id": "i39", "text": "You take your LinkedIn copy, and paste your LinkedIn bio and work experience, just like select text, copy, paste it. 'I need a website, my name is so-and-so, here's my LinkedIn history. My favorite color is blue, and I like dogs.' And then hit paste. And then you can hit deploy. And if you don't have a .com yet, now you can.", "context": "Recommended first-time user exercise: personal website from LinkedIn bio demonstrates zero-shot capability", "topic_id": "topic_19", "line_start": 698, "line_end": 704 }, { "id": "i40", "text": "I mean, now you have a real, personalized service. I think there's kind of a moment around that where it's like, 'Oh, okay, wow. This,' it's zero shot, zero shot 99.999% of the time. You're getting a beautiful personal website that you didn't have before, that would've taken you an hour, on Wix. If not more.", "context": "Personal website creation takes minutes instead of hours on traditional website builders", "topic_id": "topic_19", "line_start": 704, "line_end": 704 } ], "examples": [ { "id": "ex1", "explicit_text": "We started the company seven years ago, and were about to run out of money and shut down. But they realized the tech that they'd been building for the past seven years, called WebContainer, was perfectly suited for building AI products in a browser.", "inferred_identity": "StackBlitz", "confidence": "explicit", "tags": [ "StackBlitz", "founding story", "near-death experience", "pivoting to AI", "WebContainer technology", "seven-year bet" ], "lesson": "Demonstrates how a company's demise can be averted by recognizing that existing technology solves an emergent market need. The long technology investment (7 years) suddenly became valuable when AI code generation became viable.", "topic_id": "topic_1", "line_start": 26, "line_end": 26 }, { "id": "ex2", "explicit_text": "When we launched, the company was on the verge of going under when we launched Bolt, our company StackBlitz. We'd been around for seven years building web-based development environment stuff. And so when we launched this we were like, 'This would be amazing if this added a 100K of ARR over the next couple of months.' And what ended up happening is, in the first two months, we went from zero to 20 million of ARR.", "inferred_identity": "StackBlitz / Bolt", "confidence": "explicit", "tags": [ "StackBlitz", "Bolt", "growth metrics", "product launch", "expectations vs. reality", "20M ARR", "two months", "startup near-death", "revenue inflection" ], "lesson": "The contrast between expected (100K ARR) and actual (20M ARR) results shows how market timing and the right product can create exponential growth that far exceeds founder expectations. Expectations grounded in past patterns become irrelevant in moments of paradigm shift.", "topic_id": "topic_1", "line_start": 65, "line_end": 65 }, { "id": "ex3", "explicit_text": "I remember our dev ops engineer, he was the one who would flag me. He was like, 'Guys, we got 60K today. This is crazy.' And I was like, 'Yeah, yeah. But this is launch day.' ... And then the next day we added 80K, or whatever it was, and it just kind of kept going.", "inferred_identity": "StackBlitz team member (dev ops engineer)", "confidence": "implicit", "tags": [ "StackBlitz", "launch day", "revenue growth", "ARR accumulation", "team psychology", "tempered expectations", "daily recurring revenue", "sign of product-market fit" ], "lesson": "Shows how founder psychology differs from team psychology during exponential growth. While engineers celebrate daily wins, experienced founders remain cautious, knowing launch day spikes often decline. Sustained growth required managing both team morale and founder skepticism.", "topic_id": "topic_5", "line_start": 250, "line_end": 251 }, { "id": "ex4", "explicit_text": "One of the first people that signed up for Bolt was this guy named Paul, and he's an entrepreneur, and doesn't know how to code. Built a CRM in three weeks, that has AI built into it and Stripe for billing, et cetera. He had gotten a quote from an agency for this, it was going to be 30 grand, and take six months. He had done in three weeks, and I think he spent 300 bucks on Bolt for the thing. So it's like, this is... And he's making money off of this. This is his startup. Right?", "inferred_identity": "Paul (Bolt early user, non-technical entrepreneur)", "confidence": "implicit", "tags": [ "Paul", "non-developer", "CRM", "Stripe integration", "AI features", "cost savings", "time compression", "production software", "startup founder", "greenfield project" ], "lesson": "Demonstrates the democratization of software development. A non-technical founder built production software (with AI and payments) in 3 weeks for $300 instead of 6 months and $30K from an agency. This is a 99% cost reduction and 8x faster delivery, making entrepreneurship accessible without developers.", "topic_id": "topic_10", "line_start": 403, "line_end": 404 }, { "id": "ex5", "explicit_text": "He got a quote from an agency for this, it was going to be 30 grand, and take six months. He had done in three weeks, and I think he spent 300 bucks on Bolt for the thing.", "inferred_identity": "Paul / Agency (unnamed contractor)", "confidence": "implicit", "tags": [ "cost comparison", "agency pricing", "development speed", "outsourced vs. AI", "product development", "CRM", "Stripe", "freelance", "time-to-market" ], "lesson": "Direct comparison shows the value proposition erosion for traditional software development service providers. Agencies will face competition from AI-assisted self-service tools, challenging the business model of custom development shops.", "topic_id": "topic_10", "line_start": 404, "line_end": 404 }, { "id": "ex6", "explicit_text": "We'd gotten the idea for this, and the insight that this might be possible, because back when my co-founder and I came out to the Valley, he and I grew up down the street from each other in Chicago, we wrote code together at 13, and been building stuff ever since. And we came out to the Valley in 2012, and we just had the good fortune of bumping into Dylan Field and Evan Wallace when they were building Figma, in the early days.", "inferred_identity": "Dylan Field and Evan Wallace (Figma founders)", "confidence": "explicit", "tags": [ "Figma", "Dylan Field", "Evan Wallace", "browser technology", "WebGL", "deep technology", "inspiration", "Silicon Valley", "2012", "serendipity" ], "lesson": "Shows how exposure to paradigm-shifting technology (Figma's browser-based approach using WebGL) informed the architectural decisions at StackBlitz. Sometimes the best insights come from observing peer companies solving analogous problems in adjacent domains.", "topic_id": "topic_3", "line_start": 193, "line_end": 194 }, { "id": "ex7", "explicit_text": "Netflix, or any of these other fan companies, the first month or two is you being onboarded, to run that stuff on your computer and set up your environment.", "inferred_identity": "Netflix (and unnamed fan tech companies)", "confidence": "explicit", "tags": [ "Netflix", "onboarding", "developer experience", "large enterprise", "setup friction", "developer productivity", "environment configuration" ], "lesson": "Even at the world's best tech companies, developer environment setup is a significant friction point (1-2 months of new hire time). This validates the pain point that WebContainer solved.", "topic_id": "topic_3", "line_start": 202, "line_end": 202 }, { "id": "ex8", "explicit_text": "Cloud 9 was the first one, back in 2009 or so. The way these have always worked is that your browser's basically doing nothing, when you go to that. Every user that gets connected, there has to be a cloud VM that gets spun up for them, and then your browser's just taking your keystrokes, sending it to the server, and then sending back the results of it.", "inferred_identity": "Cloud 9 (Cloud IDE, founded 2009)", "confidence": "explicit", "tags": [ "Cloud 9", "Cloud IDE", "cloud VMs", "architecture", "server-side rendering", "browser thin client", "scalability bottleneck", "developer tools" ], "lesson": "Shows the historical architecture of cloud development environments and why they hit scale limits. The server-side compute model doesn't work beyond certain user counts because there aren't enough VMs to rent. This justified the local-compute approach of WebContainer.", "topic_id": "topic_3", "line_start": 206, "line_end": 206 }, { "id": "ex9", "explicit_text": "I think that that's the best way to put it. It's exactly that. And so it's very interesting to just... And I think as people say, 'Okay, these are just toys, they're prototypes, it's not going to work with your existing code, it's not going to scale.' It's important to note just what we talked about. This is a model from last June that this is possible on, and everybody's working on the next cutting edge model that will make this even better.", "inferred_identity": "Claude Sonnet (Anthropic model, launched June 2024)", "confidence": "explicit", "tags": [ "Claude Sonnet", "Anthropic", "LLM", "model capability", "June 2024", "code generation", "frontier model", "improvement trajectory" ], "lesson": "Emphasizes that current AI code generation limitations are based on a 9-month-old model. As new models launch, capabilities will improve dramatically. Skeptics dismissing AI tools as 'toys' are judging based on yesterday's technology.", "topic_id": "topic_16", "line_start": 608, "line_end": 608 }, { "id": "ex10", "explicit_text": "So by the time, and I'm going to go back and tell our engineers, 'I said this on this podcast-' 'I've committed. Sorry, guys.' This is, I found actually being a leaky faucet on talking on podcasts and stuff, my engineer is like, 'How could you tell them...' 'You just have to ship it faster, now. You got to make it real,' right?", "inferred_identity": "StackBlitz engineering team", "confidence": "implicit", "tags": [ "StackBlitz", "engineering team", "public commitments", "podcast announcements", "shipping pressure", "deadline creation", "team dynamics" ], "lesson": "Shows the dynamic where public statements (podcast commitments) create internal shipping pressure. Engineers use founder's public promises as deadlines to motivate faster execution. This is a common startup growth hack.", "topic_id": "topic_17", "line_start": 614, "line_end": 620 }, { "id": "ex11", "explicit_text": "Anthropic ran out of GPUs for us. Dario emailed me, he was like, 'Listen, we don't have anything more to give you.'", "inferred_identity": "Dario Amodei (Anthropic CEO)", "confidence": "explicit", "tags": [ "Dario Amodei", "Anthropic", "GPU shortage", "infrastructure constraints", "inference capacity", "supply constraints", "hypergrowth challenges" ], "lesson": "Even Anthropic (the model provider) ran out of GPU capacity to serve Bolt's inference demands. This illustrates the infrastructure stress caused by explosive API adoption and shows that supply constraints, not demand, became the bottleneck.", "topic_id": "topic_5", "line_start": 269, "line_end": 269 }, { "id": "ex12", "explicit_text": "We partnered up with a company called Expo, and their entire business is making, basically, React Native tooling and this ecosystem that makes it super easy to build beautiful apps, and actually get them in the app store.", "inferred_identity": "Expo (React Native tooling company)", "confidence": "explicit", "tags": [ "Expo", "React Native", "mobile development", "tooling", "app store deployment", "integration partner", "mobile-first" ], "lesson": "Strategic partnership between Bolt and Expo created mobile app capability. Shows how Bolt extended its reach by integrating with existing specialized platforms rather than building everything in-house.", "topic_id": "topic_2", "line_start": 131, "line_end": 131 }, { "id": "ex13", "explicit_text": "We've partnered up with a company called Anima to do this. But basically, so on any Figma URL, when you're looking at a design that you've made, if you just put bolt.new in front of that URL and hit enter, it's going to suck that design into Bolt, and turn it into a full stack app or mobile app, just out of the box.", "inferred_identity": "Anima (Figma-to-code platform)", "confidence": "explicit", "tags": [ "Anima", "Figma", "design-to-code", "integration", "automation", "no-code", "plugin ecosystem" ], "lesson": "Partnership with Anima (the leading Figma-to-code tool) integrates design directly into development. This closes a loop where designers can export Figma designs directly to functional code, enabling designers to build without developers.", "topic_id": "topic_17", "line_start": 629, "line_end": 629 }, { "id": "ex14", "explicit_text": "I mean, I think we've got a subscription to Claude, and ChatGPT, and things like that. But I think for development, Cursor is the main thing.", "inferred_identity": "Cursor (AI code editor by Anysphere)", "confidence": "explicit", "tags": [ "Cursor", "code editor", "AI tool", "engineering stack", "developer tools", "VSCode-based" ], "lesson": "Even StackBlitz engineers (building Bolt, a code generation product) prefer Cursor for coding their own codebase. This shows that context-aware, codebase-integrated AI tools win over general-purpose chat models for large codebases.", "topic_id": "topic_18", "line_start": 674, "line_end": 674 }, { "id": "ex15", "explicit_text": "My co-founder and I, we didn't go to college. My co-founder dropped out of college after a semester or something, but I didn't go, because I was like, 'We're coding,' we were doing contracting at the time, making money.", "inferred_identity": "Eric Simons (founder) and Albert Pas (co-founder)", "confidence": "explicit", "tags": [ "Eric Simons", "Albert Pas", "no college", "dropout", "self-taught", "early-stage bootstrapping", "alternative education" ], "lesson": "Both StackBlitz founders skipped college and instead focused on building and earning. This alternative path to technical expertise is becoming increasingly viable with free online resources.", "topic_id": "topic_13", "line_start": 496, "line_end": 497 }, { "id": "ex16", "explicit_text": "Google, and Open AI, all the... Everyone is now gunning for coding because, how big is the market opportunity to rewrite the software world order? It's trillions of dollars, or something, right? The world runs on software.", "inferred_identity": "Google, OpenAI (and unnamed competitors)", "confidence": "explicit", "tags": [ "Google", "OpenAI", "market competition", "coding AI", "trillion-dollar market", "AI arms race" ], "lesson": "The success of Sonnet woke up the tech industry to the coding opportunity. Every major AI company is now investing heavily in coding models because the market opportunity is massive (trillions, since software underlies all modern companies).", "topic_id": "topic_16", "line_start": 563, "line_end": 563 }, { "id": "ex17", "explicit_text": "When I first came out to Silicon Valley, we got into a, I was building a K12 education startup at the time, and this is back during the days where Y Combinator, they only gave you like 20 grand, and so there was this offshoot of Y Combinator called Imagine K12, it was Jeff Ralston who, I think he was CEO of YC a couple of years back.", "inferred_identity": "Jeff Ralston (Imagine K12 founder, former YC CEO)", "confidence": "explicit", "tags": [ "Jeff Ralston", "Imagine K12", "Y Combinator", "2012", "early accelerators", "K12 education", "startup funding" ], "lesson": "Shows the emergence of verticalized accelerators (Imagine K12 for education) and YC's evolution. Also shows the context: in 2012 YC only funded with $20K, so entrepreneurs had to bootstrap heavily or live in office spaces.", "topic_id": "topic_20", "line_start": 719, "line_end": 719 }, { "id": "ex18", "explicit_text": "So I was there, and we ended up running out of money. 20 grand doesn't go very far in the Valley, so three or four months in, we were like, 'Oh God, what do we do?' And I was going to the AOL office multiple times a week, because we had access cards to get in to get to the investors' offices. And I realized, I was like, 'You know, they have couches here, and they have food. There's ramen that you can microwave, and there's a gym where there's a shower, and even you can do laundry.' And then, so I was like, 'I don't know, maybe while I figure this out, I'll just live out of here.'", "inferred_identity": "AOL (headquarters in Palo Alto), Imagine K12 (startup accelerator)", "confidence": "explicit", "tags": [ "AOL", "Palo Alto", "squatting", "office living", "resourcefulness", "startup desperation", "survival tactics", "2012" ], "lesson": "Demonstrates extreme resourcefulness: when money ran out, Eric leveraged free resources (food, shower, gym, couch) at AOL's office. This shows the depth of hustle required in pre-product-market-fit startups and willingness to live unconventionally.", "topic_id": "topic_20", "line_start": 722, "line_end": 722 }, { "id": "ex19", "explicit_text": "And so I was there for, I think, four or five months. I was living out of this headquarters over on Page Mill in El Camino, in Palo Alto. And then, I got away with it for a while just because the guards, the security guards, they worked 12-hour shifts. And so the guys that, when I was there at night... And I was coding, all day every day, basically. So the guys at night just were like, 'Dang, this guy works really hard.' And then in the morning they'd be like, 'Wow, this guy is working really hard.' And I became friends with some of them, and then eventually, I think there were also a whole bunch of Stanford students that I think they put bunks in one of the aisles.", "inferred_identity": "Security guards at AOL, Stanford students", "confidence": "implicit", "tags": [ "AOL office", "security guards", "Stanford", "living in office", "code marathon", "community of founders", "Page Mill Road" ], "lesson": "Shows how security guards became allies through relationship-building and perceived work ethic. Also reveals that office squatting was a widespread phenomenon (Stanford students also bunking in aisles), suggesting a cohort of founders surviving on minimal resources.", "topic_id": "topic_20", "line_start": 722, "line_end": 725 }, { "id": "ex20", "explicit_text": "And then one morning, at like 4:00 in the morning, a guard came in and threw me out. I'm from Chicago. I don't know anyone. At that point I'm like, 'I know no one in the Bay Area.' So I went to a Starbucks, which was not open. I slept on the table outside of the thing. And I think I hit up one of the other entrepreneurs that was in the program. I was like, 'Do you have a couch? I think I kind of need it, at this point.'", "inferred_identity": "Imagine K12 entrepreneur community", "confidence": "implicit", "tags": [ "AOL eviction", "homelessness", "couch surfing", "Bay Area", "founder network", "desperation", "community support" ], "lesson": "Shows the low point: evicted at 4 AM, sleeping outside Starbucks, asking fellow accelerator participants for a couch. Demonstrates how founder networks provide safety nets for those in extreme situations. Also shows Eric's vulnerability and willingness to ask for help.", "topic_id": "topic_20", "line_start": 725, "line_end": 728 }, { "id": "ex21", "explicit_text": "Yeah, the press got wind of it, and it was this worldwide story. But I lived on a dollar a day. That was the crazy thing. My burn rate was a dollar a day, at that time.", "inferred_identity": "Eric Simons, media outlets", "confidence": "explicit", "tags": [ "Eric Simons", "press coverage", "viral story", "$1/day", "McDonald's Dollar Menu", "extreme frugality", "publicity" ], "lesson": "The AOL squatting story became international news. Eric's extreme frugality ($1/day) during this period shows the mindset that later allowed StackBlitz to bootstrap for years with minimal burn. Poverty-driven resourcefulness became a company value.", "topic_id": "topic_20", "line_start": 728, "line_end": 728 }, { "id": "ex22", "explicit_text": "This is back when McDonald's had the Dollar Menu. Literally. So it was like, I occasionally would go and get a cheeseburger or whatever. Yeah, it was ultimate scrappiness.", "inferred_identity": "McDonald's (Dollar Menu era, pre-2013)", "confidence": "explicit", "tags": [ "McDonald's", "Dollar Menu", "budget living", "survival food", "2012", "frugality hack" ], "lesson": "Shows how product availability (Dollar Menu) enabled survival at extreme poverty. This is a specific historical fact (Dollar Menu was discontinued): Eric's story is time-bound to 2012 when this was possible.", "topic_id": "topic_20", "line_start": 734, "line_end": 734 } ] }