What the people actually building AI said today. One page — a 2-min read.
Today’s biggest bets were on how AI learns and how safely it acts. One lab laid out a path beyond language models, while product teams shipped new agent capabilities and builders warned that old security practices are running out of time.
David Silver bets AI can discover knowledge without human training data
Ineffable Intelligence is building agents that learn through experience in simulated worlds, aiming for discoveries beyond what human data can teach them. Silver expects strong evidence for the approach within a year and now forecasts a superintelligence in about three years. “There’s this other lamppost that right now no one is looking under.”
Claude brings its full platform to Google Cloud
Anthropic-managed Claude Models on Google Cloud are generally available, with Google Cloud credentials and billing. Fast mode for Opus 5.5, code execution, and the Skills API are available there today. Next week, Claude sessions on Pro, Max, and Team plans will default to auto mode, with an automated check blocking risky actions.
Decisions API reaches general availability
Thibault Sottiaux says the Decisions API is generally available and OpenResponses has joined the Linux Foundation. He also reports performance improvements and says Uber is going all-in on Codex and Work.
Aaron Levie says agents end security through obscurity
Autonomous systems can probe every corner of legacy infrastructure and try attack after attack, Levie argues. He expects companies to modernize access controls, clean up data pipelines, and use agents to simulate risks. That work is one reason he thinks AI adoption will take longer than some expect.
Guillermo Rauch calls for an end to static credentials
Rauch wants stolen credentials to expire quickly, limiting what attackers can do with them. He calls for an industry push around OIDC comparable to the adoption of website encryption.
Robotics is moving toward plain-language instruction
Aditya Agarwal says robotics is following a shift seen in language models: from fine-tuning to prompting to simply describing the desired result. He points to Physical Intelligence’s π0.7 learning new skills from plain-language coaching.
Nikunj Kothari sees a sharper squeeze between seed and Series A
Kothari says nearly 10% of the seed companies he meets are acquired within months of their first meeting. His read is that starting a company has become easier while scaling one has become harder, and he expects the squeeze to persist.
AI agents need to estimate work on their own clock
Swyx calls Opus 5.5 the best work estimator model he has used. As teams delegate more work, he argues agents should estimate their own parallel execution time instead of repeating a human estimate of days when they might finish in hours.
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.
