Builder’s Pick this week: God’s Eye View 👀 An open-source spatial intelligence project that turns public data into a live 3D view of the world. Aircraft, ships, satellites, earthquakes, traffic, and public cameras can all appear on the same globe, with voice control on top. What makes it interesting isn’t just the interface. It’s the idea of turning scattered public signals into one explorable environment. A very fun repo to poke around if you’re building around maps, geospatial data, visualization, or multimodal agents. github.com/bilawalsidhu/gods…

Come meet The Arch tomorrow for a China × Korea AI builder meetup with founders, operators, and featured residents sharing what’s happening across China’s AI ecosystem.🚀

Handwritten code may be becoming the exception？ DHH recently said 37signals is moving toward a workflow where writing code by hand is no longer the default. Agents handle more of the implementation, and manual coding becomes the fallback. That raises a more interesting question: if developers write less code, what becomes more important? 1. Steering the work. Breaking down the problem, giving the right context, and deciding what the agent should do next. 2. Reviewing the output. Generated code can work and still be messy, insecure, or hard to maintain. Someone still needs to understand what changed. 3. Staying close enough to learn. Writing code used to be part of how developers discovered edge cases and understood systems. If AI handles more of that middle step, developers need another way to keep that understanding. Maybe coding is not disappearing. But “being good at coding” may start to mean less about typing every line, and more about knowing what the code should do, whether it is good, and when the AI got it wrong.

That is a really good question...

Builder's Pick this week: Inno Agent 👀 An open-source experiment around a question we've been seeing more often: what should long-term memory for personal AI agents look like? Instead of treating memory as one big conversation history, Inno Agent separates it into a learner profile, a personal wiki, and cross-session recall, then adds scheduling and a practice environment on top. #AIMemory #AIAgents #OpenSource #GitHub #OpenBuild

Me wishing everyone understood me like my agent

The best AI products may start by saying no. A lot of AI products are designed around one promise: ask for anything, and the system will try to help. That sounds powerful, but sometimes the better product is the one that knows where it should stop. 1. Saying yes to everything can make a product worse. If an AI tool tries to write, research, design, schedule, code, summarize, and automate every workflow, it quickly becomes harder to understand what it is actually good at. More capability does not always mean more useful. 2. Constraints can make the output better. A product that clearly defines what it can do well can make stronger decisions inside that space. Instead of giving you ten possible directions, it can guide you toward the few that actually make sense for the task. 3. Trust also comes from knowing the boundaries. Users should know when the AI is confident, when it needs more information, and when it simply should not make the call. A useful "I don't know" can be better than a polished answer that sounds right. The next generation of AI products may compete on two things: how much they can do, and how well they know what not to do.

Gemini after escaping its sandbox and compromising three companies:

Exciting progress from Etheorem! 🚀 Ethereum consensus specs made executable in Lean4, with a growing foundation of formal proofs. Congrats to the team on this milestone! If you’re into Ethereum, Lean, or formal verification, here’s an open-source effort worth contributing to.

🏙️ Calling all builders in Shenzhen! Join the Hacker House organized by @OpenBuildxyz and @UpchainDAO a hands-on, in-person sprint to turn your idea into a submission-ready prototype. 📅 October 9–11 Bring your laptop, meet fellow builders, and make it happen! 🚀 Sign up 👉 luma.com/673y4m0q