Anthropic published the complete origin story of Claude Code today, told by core team members Boris Cherny, Cat Wu, Thariq. The official tweet reached 2,959 likes and 314 retweets. Boris emphasized: "So much more to do. We are 1% done."

Replit CEO Amjad Masad announced the agent self-improving loop is closed. A viral customer story: an Atlanta real estate firm replaced Salesforce with a Replit-built CRM, saving USD 100k/year. The tweet received 1,487 likes and 92 retweets.

Vercel CEO Guillermo Rauch asks: Is software getting better? My ability to influence productivity has radically increased. I feel agency and autonomy I never had before. Personal software is real. At Vercel, our CTO built industry-impacting software nearly autonomously with agents. PMs ship software, juniors and interns ship quality code at scale. But the ultimate jury is whether users love the product, expand usage, and refer friends. That is the main metric he watches.

Zara Zhang recommended three must-watch AI conference talk collections: AI Engineer Summit, Cursor Compile, and Figma Config. She noted these high-quality talks are freely available online (conference tickets often cost hundreds of dollars), yet most videos only get tens of thousands of views. Watching on YouTube is arguably better than in-person: no bad audio, no crowd blocking the screen, no distractions, and you can pause to take notes. 535 likes, 68 retweets.

Claude Code now supports Artifacts — capturing work progress as live, shareable visual pages. Built from session context automatically with no data wiring needed. Pages update in real-time with version history. Available in beta for Team and Enterprise orgs. Key use cases: incident timelines, security findings, cloud cost maps, PR walkthroughs, UX variants, onboarding reports. All pages default to private, viewable only by authenticated org members.

Edwin frames AI training as "building a school for AGI" — models arrive unformed like children and must learn everything from basic arithmetic to taste, poetry and beauty. Surge reached about USD 1 billion in revenue without VC funding. Topics: math capability leap from GSM8K to Riemann, AI should optimize for human growth not engagement, data environments as the new training paradigm, the delegation vs. autonomy balance, and AGI timeline (Fields Medal/Nobel level within 5 years).

Dan Shipper raised the "delegation paradox": the more you delegate, the more human capability atrophies — like driving instead of walking causing muscle loss. Edwin acknowledges there is a time for both, but warns against mindless delegation. He quotes Ted Chiang: "You have to pretend that you have free will. We must consciously choose to prove things on our own, to write on our own, to create on our own — because preserving our humanity is valuable in and of itself, even if the output is not optimal."

Edwin contrasts two product paths: one optimized for "keeping you on the site" like certain social media; another like what Facebook could have been — recommending restaurants and movies to enjoy with friends, encouraging real-world connections. AI faces the same choice: become another addictive social media, or become a tool that helps you grow and encourages you to create yourself. He warns about internal tension at AI companies: researchers want to advance capability, PMs need measurable metrics. "Session length" and "users" are easy to measure; "improving human lives" is much harder and longer-term.