·
Finding signal on Twitter is more difficult than it used to be. We curate the best tweets on topics like AI, startups, and product development every weekday so you can focus on what matters.
PLUS: Stripe and Meta back agent standard
OpenAI released a GitHub repository with over 700 math problems solved by its models.
Stripe, Meta, Shopify, and Walmart are founding members of a Personal Agent Protocol for how agents interact with businesses.
OpenAI's math post is live on GitHub, after Wired reported a mass release was coming.
Per Wired, OpenAI planned to post solutions to hundreds of unsolved math problems on GitHub in one mass release.
Many papers appeared the night before, as people tried to publish ahead of OpenAI's reported GitHub release of newly solved problems.
OpenAI staff say their models solved important math problems, documented in a new blog post.
OpenAI's math release follows months of mathematicians asking for proofs, not blog posts, according to this timeline of the dispute.
Stripe says Meta, Sierra, Genesys, Instinct, Rocket, Shopify, and Walmart are founding partners of an open standard for personal agents.
Google's EmbeddingGemma 2 puts text, code, images, video, and audio into one embedding space under an Apache 2.0 license.
OpenAI's Decisions API is live, and a staffer says a realtime general classifier opens new possibilities.
OpenAI put 25% of its production engineers on defense, using its models to find vulnerabilities in its own systems.
Anthropic announced that Claude now works inside Google Docs, Sheets, and Slides.
Anthropic's announcement explains screenshots of people talking to Mythos 5.1 that circulated earlier in the morning.
SpaceXAI's Grok 4.7 is now live on Microsoft Foundry.
Grok 4.7 scores higher than Grok 4.6 on ARC-AGI-1, but lower on ARC-AGI-2 and ARC-AGI-3.
One engineer automated releases and QA with Grok Bot, running 10+ agents that click around each build like real users.
ChatGPT leads consumer AI, at 2x Gemini's scale and 6x Claude's on web, according to a16z data.
Mistral is the best open-weights model from the US or Europe on aggregated benchmarks, after raising €3 billion.
Patreon's CTO and its heads of product and engineering are leaving to lead OpenAI's new creator division.
Every launched a Slack agent that lets coworkers see how AI work gets done instead of only the finished result.
Ampersand raised a $15M Series A from Bessemer on the idea that custom CRM fields can become configuration.
Cloudflare shipped 46 announcements in one week for its 2026 Birthday Week.
Claude Opus 5.5 generated 500 canvas visualizations in vanilla JavaScript, with no Three.js, WebGL, or other libraries.
Dario Amodei looked at GPT-2 in 2019 and saw eight more zeros of compute as a path to AGI, per Kevin Roose.
Kevin Roose says a slow accumulation of distrust between Amodei, Altman, and Brockman led to Anthropic.