Golf (YC X25)

Everyone's doing 996. We're doing 996S. During YC, me and Antoni Gmitruk were just grinding. All-nighters became normal. We barely left the apartment. Thought that's what it takes to win. We were shipping fast but thinking slow. Decisions that should take 30 minutes took 3 hours. Everything felt harder than it should. Then Brian Chesky said something that hit me: "Just do your daily walk. Think through things while you walk." And Aaron Epstein: "30 minutes a day won't kill your startup. Lack of clarity will." I started small. Morning walks. Then runs. Then I got back on my bike. Now I do this every single day: Gravel biking every other day (did 70km this Saturday) Gym on off days - just lift heavy shit 10,000 steps minimum - I take calls while walking Actually sleep and track recovery with Whoop The difference is insane. Problems that felt impossible at midnight are obvious after a morning ride. Code reviews I'd agonize over for hours take 20 minutes with a clear head. The S in 996S? Sport. Sanity. Stamina. You can't build for 10 years if you burn out in year 1. Your startup needs you healthy more than it needs you at your desk 16 hours a day with a fried brain. Turns out, a healthy founder is a high-performing founder. Who knew? Take care of yourself. This is a marathon, not a sprint. And frankly, it makes the whole grind a lot more enjoyable. ps.: follow me on strava https://lnkd.in/d99d53xV if you want to see me suffer on a bike:)

Everyone's worried about the wrong MCP security threats. Tool poisoning? Rug pulls? Tool shadowing? If you're running enterprise MCP servers with basic security hygiene, these don't matter. What will actually get you breached: Authentication misconfiguration (OAuth done wrong) Indirect prompt injection Wrote a no-bullshit breakdown of what's real vs what's hype, and how to prioritize if you only have 1 week. Read: https://lnkd.in/dASsEPw7

Golf (YC X25) is coming to London for MCP Dev Summit tomorrow! If you're building MCP servers for enterprises and want to prevent agents from going rogue, hit me up. Let's grab coffee and talk security!

Your AI agent can say “Let me check on that.” Now it’s time to make it actually check. Join us for our next workshop where we'll be sitting down with Golf (YC X25) CEO Wojciech Błaszak to talk through what MCP (Model Context Protocol) actually is, how companies are using it, and where this is all going. We’ll cover: 1. How MCP enables secure, flexible agent workflows across tools 2. Lessons from building real-world integrations at scale 3. The future of agentic platforms, and what it means for support Perfect for: AI builders, Support teams, ops leaders, founders, and anyone designing tool-augmented support systems. Speakers: Wojciech Błaszak (CEO Golf (YC X25) ), Marie Schneegans (CEO 14.ai ), and Maya C. (Founding Engineer 14.ai ) RVSP here: https://lnkd.in/edhwvUwd

have you seen our OSS framework golfMCP? suuper easy to build out MCP server with auth, telemetry, debugging and deployment. The only thing you should focus is building great logic - no more plumbing. take a look and leave a star here: https://lnkd.in/dww-xrvr

We got to the Y Combinator spring batch! Golf (YC X25) Antoni Gmitruk and I applied to YC for the first time when we were 17, from there we needed 3 attempts to get in. Feels like the best 0 point to be, you don't have to worry about anything else than building sth people want - just execution. Finally, we found our community where we feel... normal? Really strange feeling. As I wrote some time ago, the current internet wasn't built for AI agents. That's why we are doubling down on building the new one with MCP. With Golf MCP gateway you can move from local to remote servers, and stay in control over deployment. Companies use Golf to deploy remote MCP servers, with full observability layer on top. If you are thinking about starting to serve AI agents as your new type of users, let's jump on a call! MCP is THE agentic interface. --- Huge thanks to our early believers for the support from day -1 !!! Tomasz Karwatka Piotr Karwatka Bartek Pucek Mati Staniszewski Tytus Cytowski Przemek Praszczalek

We just launched v0 of Authed! (check it out!) Today's AI agents are forced to impersonate humans - browsing through UIs, handling password flows, and waiting for human approval. But as models improve and agents become more autonomous, we deeply believe the future isn't about mimicking human behavior. It's about building optimized infrastructure specifically designed for AI agents. (imagine milions of agents rendering pages to make "transaction" ;)) That's why we're taking the first step with Authed - treating agents as legitimate users across the web (we can think of them as premium traffic). Our v0 release allows you to register agents with unique identities and enable them to interact securely with other agents, without static credentials or human bottlenecks. Can't wait to see new verticals built on top of agent-to-agent interactions! The machine-to-machine internet needs its own authentication layer, and we're building it. Go try it out, give us feedback, contribute - let's start building!