What the people actually building AI said today. One page — a 2-min read.
The clearest theme today was who controls the tools around AI agents. Builders shipped ways to put agents to work, while Modal argued that the infrastructure around model calls may prove more valuable than the calls themselves.
Modal says a full AI platform is a stronger bet than serving tokens alone
Modal’s Eric argues that basic model serving is becoming a race to the bottom. “Customers, they don't wanna buy, like, single trick ponies.” Modal has launched dedicated model inference options alongside its sandboxes and training tools; he says sandboxes now account for about a third of its revenue.
Karpathy built a YouTube alternative from 20 years of Hacker News comments
Andrej Karpathy used Claude to sift through 43 million Hacker News comments and surface 6,454 repeatedly recommended videos for HackerTube. He says the project took about two hours, with Claude handling the code and deployment after he supplied an API key, server, and domain. His broader point: projects that once took weeks or months can now be built in a morning.
Replit can turn site feedback into prompts for building agents
Amjad Masad says Replit users can enable a feedback widget with one toggle, turning visitors’ comments into prompts for agents to build from. He also predicts that subsidized AI subscriptions will become more expensive or disappear.
Levie sees knowledge workers becoming managers of agents
Aaron Levie argues that agents will shift people from doing tasks with AI assistance to directing teams of agents. He expects the biggest effects in information-heavy fields such as coding, legal work, cybersecurity, and biotech, where output could become less tied to hours worked.
Rauch urges builders to own their agent tools and data
Guillermo Rauch argues for keeping control of an agent’s data, connections, and runtime while maintaining access to different models. He wants local models and open source to keep growing, even though the largest models remain out of reach for local use.
Peter Yang sees no single AI tool winning every task
Peter Yang says his current picks differ by job: ChatGPT and Codex for browser and computer use, Claude for design and video editing, and Grok Bot for continuous agents and scheduled jobs. He also shares a Google Labs product director’s question about whether some products need people to interact with their interfaces at all.
Source data comes from the open-source project follow-builders by zarazhangrui, released under the MIT license. Summaries are generated by an LLM from that project's public feeds, and the summarization prompts are adapted from it. Every item above links to its original source.
Summaries generated automatically. Read the original before relying on any claim.
