Inside @Kalshi : How a Prediction Market Hit $22B in 2 Years 65 lawyers said it couldn't be done. They sued their own regulator instead. Eight years later, Kalshi just raised $1B at a $22B valuation, and co-founders @mansourtarek_ and @luanalopeslara took us inside the bet that got them here. 00:00 Why prediction markets? 01:35 We sue our regulator 02:31 From Ballet to Building a $22B Company 03:19 Discipline & delayed gratification 04:20 Getting out of a comfort zone 04:56 65 Lawyers said No 05:05 Tarek's story 06:04 How the idea starts - The Trump trade 07:45 No one says it's possible 09:00 Four Years to Earn the Right to Launch 10:22 Stagnating while everyone else grew 11:53 Cutting through the noise 12:43 "Forget product-market fit" 13:28 Why we sued the regulator 14:45 We won. Then they appealed 16:43 The 4-week sprint to the election 18:03 The clearinghouse crisis 18:42 6 months of migration in one weekend 19:14 Calling the election before the media 19:31 Prediction markets vs gambling 20:51 Money where your mouth is 22:06 What "regulated" actually means 23:10 It's not just sports 24:05 The trader who studies Trump's speeches 26:40 The teacher who made $8K from source code 27:42 Hurricane markets as insurance 28:40 Where Kalshi goes next

Mike Schroepfer, former Meta CTO, now founder of Gigascale Capital: "How far away from theoretical maximum is the current version of the thing?" In a 21-minute talk, he gives the first test he runs on any new technology: how much room is left to scale. Most people bet on what already works. The few who compound bet on what still has thousands of times of headroom. At 99% of the speed of light there is nowhere to go. At 0.001%, the whole runway is yours. Bookmark this.

The story of the little building where Silicon Valley's legends were born Saeed Amidi came from a wealthy Iranian family. He moved to the US to study business. Then the Iranian Revolution severed ties between the US and Iran. The money from home stopped overnight. He had to start a business fast. So he opened a small office in Palo Alto and started a packaging and bottled water company. The business took off, and he started making money. A tenant himself, he eventually bought the building. Then he began renting out the spare office space to small tech companies. One of them was Google. It moved in with 6 employees and left with around 60. Watching it grow got him interested in tech. Then PayPal. Amidi negotiated to take part of the rent in equity instead of cash. That made him one of its earliest investors. Danger, Milo, and WePay all passed through this building too. Danger was acquired by Microsoft for around $500M, Milo by eBay for $75M, and WePay by J.P. Morgan for around $300M. As these successes piled up, people started calling this little building "The Lucky Building" He realized the building's real value wasn't the rent. It was the companies growing inside it. So he founded Plug and Play to back early-stage startups. Plug and Play became one of the world's most active accelerators. Today it has invested in over 2,000 companies across more than 60 locations worldwide. Dropbox, LendingClub, and Honey all came through it.

How does Figma survive the Claude Design crisis? That was the question I had while watching the Config 2026 keynote. The world had shifted a lot in a single year. Claude Code and OpenClaw had arrived, and then Claude went one step further and shipped Claude Design. A designer I know actually let out a small, wistful sigh about it. "Feels like we're heading into an era where you don't even need to open Figma anymore." But, strangely, watching Config 2026, I came away with the opposite feeling. That Figma had an edge from the very beginning, and they know it. The first thing they unveiled at Config was Code Layers. In a way, it was the obvious move. To hold off tools like Claude Code and Lovable creeping into design's territory, Figma had little choice but to take code in. What stayed with me, though, was a line from Nikolas, the product manager who introduced it. He said that while agents have made writing thousands of lines of code easy, the workflow around code was never designed for exploring together. When you work through a chatbot-style agent, everyone just lobs their own prototype at each other. Nobody is actually building together. A fair point. And his answer to that turned out to be something deeply familiar. Figma's canvas. A place to explore ideas, compare them, and build side by side. The remarkable part is that Figma has had, from the very start, the one space a chatbot interface can't be. It isn't even a new bet. It's the thing Figma has always been. Around the keynote, Dylan said AI had lowered the floor but he wasn't sure it had raised the ceiling. He also said the ceiling, in the end, is you. So here's the thing I keep turning over. If we're entering an era where one person directing a fleet of AI agents can outproduce a whole team, then how high could those people climb inside a system actually built for them to climb together? Since ChatGPT 3.5, we still haven't really escaped the same bland screen with a single search box in the middle. Which makes me wonder. The products that already had a fundamentally different shape, the ones that bolt AI on fast, might be the ones that quietly win the application layer. @figma @zoink @nikolasklein

RoboStrategy CEO @Rewkang on the future of open source AI: "One of my views is that open source models are going to get really good." "2–3 years ago, open source was less than a few percent of all tokens produced. Now it's 25–30%, maybe more." "Even if a gap remains between open source and frontier, that gap may not matter for most tasks." "As long as these open source models get to that level, the model layer will almost commoditize for physical AI, maybe in the next 3 to 5 years. At that point, intelligence becomes really cheap." "The most valuable companies are probably going to be the ones that are doing deployments. They're producing the hardware, or innovating on new designs or components." Full interview coming soon on @eo_studio @RoboStrategy

Christopher Pedregal(@cjpedregal) sold his last startup to Google. Then he built granola(@meetgranola) into a $1.5B unicorn. Our full interview, on building products people actually use: 0:00 Intro 1:42 Build the User's Best Friend 5:50 The 50% Rule: Cut your product in half before Day 1 9:29 Systemize the Feedback Loop 11:16 Stop Collecting Redundant Feedback 11:43 Be Skeptical, Critical, and Honest 12:19 Use Conservative Metrics 13:21 Trust Your Intuition Over Feedback 14:35 Shorten Feedback Loops

The more widespread tech becomes, the more "humanness" they need to be. While @AnthropicAI is continuously develops it's frontier models, they're also hiring - Event marketers - Field marketers - Community builders only 0.3% of AI users actually PAY to use the frontier models. For normal non-tech savvy people like me, the trust, the vibe, and real conversation with the people behind is more important. "humanness" is the new moat.

A lot of young SF founders say they can replace Gong with an AI-native approach. So I asked Amit @banditmove, @Gong_io's co-founder and CEO, what he thinks. He welcomed the competition. Here's his answer: Writing the code is one part of a software company. Actually running it, supporting production, driving adoption, that's extremely hard. There's a security vulnerability you need to patch in a library at 3am. What do you do? When you're alone, nobody cares. Once you're ahead, there's a crosshair on your back. We've built this system over 10 years. There's a lot of complexity Gong is still growing well, and applying AI very fast. And all founders that I know were building Gong's replacement, they pivoted to other items. Replacing a well-established OG product with AI is way harder than we think right now. Over half the fastest-growing startups in SF that I know still hire tons of sales experts. They build real human relationships and use AI to get better at the work, not to replace all of it with agents.

There are two groups of people in YC. Young founder hackers: 18 to early 20s. Great at building product. Strong backgrounds like top colleges, student founder experience, GitHub stars, etc. Native English speakers. You don't need to check every box though. In interviews, YC partners want to see your potential, not just your current status. Immigrant national champions: mid 20s to 30s. Exceptionally good at building product. Already built a great track record in their home countries. Not fluent in English. If they raised in their home country they could get at least 2x more money than from YC. But they give that up to compete as a world champion. With this group, YC partners usually dig into their numbers and their commitment to the US market, because they've seen so many immigrant founders give up the American dream and go back home. I've seen so many young founders learning product skills and insights from the second group. And the second group learning US culture and perspective from the first group.

we're iterating dude