Anthropic dropped Sonnet 5, the government held OpenAI’s latest releases back, and a Chinese lab released what Databricks’ Yuchen Jin called “the open-source Claude moment.” We chatted with 3 AI CEOs about the state of AI models in Summer 2026, and why the labs are now chasing “good enough” over “smartest.” Plus we got everyone’s AGI timeline and P(doom) scores. 🎥 Watch the full episode here 👇 0:00 Making sense of US AI policy 5:09 Tiny Action Models vs. LLMs 7:02 Runlayer, AGI Inc, and computer-use agents 10:05 What does "trusted access" really mean? 12:37 Why Krea open sourced its image model 22:21 BREAKING: Claude Sonnet 5 is here 26:00 GLM-5.2 is the open source Claude moment 31:58 Personalization and corporate ontology 41:25 Can we trust Higgsfield's numbers? 50:39 Distillation: The threat is real 58:01 Agentic personhood and corporate rights 1:03:19 Everyone's AGI timeline and p(doom) score cc: @viccpoes, @krea_ai, @divgarg, @AGI_Inc, @berman66, @runlayer, @alex, @jason

Is this the weirdest VC story out there? They asked them for a meeting. The investor came late. Sweaty. In workout clothes. This wasn't the worst. He kept shaking his metal protein shaker throughout the whole pitch. You can't make this up. @dsa

Apple hasn't fixed this problem since 2011. Using Siri is slower and more painful than using your browser. The UX isn't clear. Voice AI does the job that Siri was supposed to do: Tackle your tasks and have human-like interactions. Siri still feels like a transaction. @dsa @livekit

Palantir published a 9-point manifesto on "AI sovereignty" this week. Jason called it back in May, almost word for word. The diagnosis is identical: "Whoever controls the model controls the outcome." Thread with the receipts 👇 x.com/twistartups/status/207…

"We are not too scared of the distillation that may happen in our model. We would encourage people to distill it if they want to. The places you can get through distillation are always below the places you can get through owning." In a moment when Anthropic's high key stressed about getting distilled, here's a counter-intuitive take from Victor Perez (@viccpoes) of Krea AI. TWiAI Episode 20 (!) drops tomorrow, also featuring @divgarg and @berman66!

This tool lets you find things you didn't even know you were looking for. Most research tools can do keyword searches. NotebookLM's agent finds topics and knows your blindspot from weeks of context. How they did it? They combined 3 agents into 1 super agent. This is a researcher’s dream. @stevenbjohnson

You wouldn't ask your tutor to write your papers. Why are you asking AI? If you use AI to think for you is like hiring a personal trainer to do your workouts. By using AI to avoid thinking, you skip the learning process entirely. Use AI as your editor or researcher to bounce ideas off or find blindspots. This is not cheating. @stevenbjohnson