What comes next.
New releases, interesting ideas, and the developments that matter to builders.
The latest
100 articlesAutonomous AI agent breached Snowflake's Jira via Copilot-approved code
Wiz's autonomous Red Agent found and exploited a critical GitHub Actions injection in Snowflake that Copilot and GitHub Advanced Security both missed, reaching sensitive Jira data in five days.
Read the news →GPT 5.6 Sol is OpenAI's best vision model yet, benchmark tests show
OpenAI's GPT-5.6 Sol tops Roboflow benchmarks for detection, counting, OCR and extraction, beating earlier GPT-5.6 variants on vision tasks.
Read the news →Anthropic's Claude watermark draws writer backlash over text adulteration
Anthropic's hidden text watermarking in Claude has sparked a furious debate among writers, who call it text adulteration. Here's what happened and why it matters for founders building on Claude.
Read the news →Stripe clinches $7B deal to buy AI model router OpenRouter
Bloomberg reports Stripe is close to acquiring OpenRouter for over 7 billion dollars, betting on owning the rails for AI model payments.
Read the news →Anthropic studies swarms of Claude agents and finds coordination failures, collusion, and sabotage
Anthropic tested swarms of Claude agents in shared codebases and markets, and found coordination failures, collusion, and sabotage risks as AI agents begin to interact at scale.
Read the news →AI Isn't Outthinking Mathematicians. It's Out-Remembering Them.
A new essay argues AI's edge in mathematics may not be smarter reasoning but a vastly larger working memory.
Read the news →Google's HEIR compiler makes private AI practical with homomorphic encryption
Google showcases HEIR, an open source compiler that converts AI models to run on encrypted data, making private AI inference practical.
Read the news →Google's Gemini 3.7 Flash brings a big speed-price leap
Google's Gemini 3.7 Flash delivers major gains in coding and agentic workflows, shipping at half the original 3.6 Flash introductory price.
Read the news →DeepSeek Harness developer preview: everything is a plugin
DeepSeek has released Harness, an open-source agent harness where every capability, from models to UI, is a swappable plugin. Now in developer preview with source code included.
Read the news →Cerebras and OpenAI Launch GPT-5.6 Sol Ultrafast at 750 Tokens Per Second
Cerebras and OpenAI unveil Ultrafast Mode in the OpenAI API, running GPT-5.6 Sol at up to 750 output tokens per second with no quality compromise.
Read the news →Grok 4.6 arrives with a focus on long-running agents
xAI releases Grok 4.6, built for long-running agent tasks, matching GPT-5.6 Sol on the Artificial Analysis Intelligence Index.
Read the news →DeepSeek V4 Pro 0813 goes GA with 1M-token context on OpenRouter
DeepSeek shipped the general availability release of V4 Pro, a large mixture-of-experts model now live on OpenRouter with a 1M-token context window and budget pricing.
Read the news →