OpenAI is shutting down Sora and folding its science team, with CPO Kevin Weil and Sora lead Bill Peebles departing as the company doubles down on enterprise AI. This is a deliberate strategic contraction — OpenAI is trading moonshot optionality for revenue-focused execution. The talent and IP exiting opens immediate opportunities in video generation and AI-for-science.
Cursor is reportedly raising $2B+ at a $50B valuation led by a16z and Thrive, driven by surging enterprise adoption of AI coding tools. At $50B, Cursor is pricing in category dominance of developer tooling — a bet that AI-native IDEs replace traditional editors across engineering orgs. This is one of the fastest valuation ascents in enterprise software history.
OpenAI's updated Codex app for macOS and Windows now includes computer use, in-app browsing, image generation, memory, and plugins — effectively transforming it from a code completion tool into a general-purpose autonomous developer agent. This is OpenAI's most direct competitive move against Cursor, Devin, and the broader agentic coding market. The 1552 HN score reflects that the developer community sees this as a genuine step-change, not an incremental update.
Google is shipping 'Skills' in Chrome — a feature that lets users save AI prompts as reusable one-click tools directly in the browser. This embeds AI utility at the browser level, potentially displacing the dozens of prompt-management and AI-overlay extensions that currently monetize this workflow. With Chrome's ~3B user install base, even modest adoption makes this a category-defining distribution move.
Cerebras is moving toward a public offering after securing an AWS integration deal and a reported $10B+ agreement with OpenAI. This signals that non-NVIDIA compute infrastructure is graduating from niche to institutional-grade. A Cerebras IPO would give the market a liquid benchmark for alternative AI chip valuations.
Simon Willison upgraded his Claude Token Counter tool to support cross-model tokenizer comparisons, revealing that Claude Opus 4.7 is the first Claude model to ship a new tokenizer. Tokenizer changes have direct cost and compatibility implications for anyone running Claude in production — different token counts mean different pricing and context window behavior. This is a low-visibility but high-impact infrastructure change that most teams will discover only when billing surprises them.
OpenAI updated its Agents SDK with native sandbox execution and a model-native harness for building secure, long-running agents that operate across files and tools. The low HN score (23) suggests this is more incremental plumbing than breakthrough — but native sandboxing is a meaningful security primitive that removes a major blocker for enterprise agent deployment. This is infrastructure work that makes the Codex and broader OpenAI agent ecosystem safer to run in production.
Anthropic published updated system prompts for Claude Opus 4.7, and Willison's diff analysis surfaces meaningful behavioral shifts in how the model is instructed to operate. Anthropic remains the only major lab to publicly document its user-facing system prompts, making this a rare window into how frontier labs are shaping model behavior at the instruction level. Shifts here often predict capability and policy changes that surface in the API weeks later.
OpenAI launched GPT-Rosalind, a frontier reasoning model purpose-built for drug discovery, genomics analysis, protein reasoning, and scientific workflows. This is OpenAI's first publicly named vertical-specific model, signaling a deliberate move to own AI-for-science as a category distinct from general-purpose assistants. Named after Rosalind Franklin, the model represents OpenAI's answer to DeepMind's AlphaFold lineage and a direct bid for pharma/biotech enterprise contracts.
That's today's briefing.
Get it in your inbox every morning — free.
Help us improve AI in News
Got a suggestion, bug report, or question?