What comes next.
New releases, interesting ideas, and the developments that matter to builders.
The latest
100 articlesClaude Code Team Moves Code Review to Machines as PRs Double
Anthropic's Claude Code team says agents now do most of the reading, with a classifier deciding who, if anyone, has to approve a change.
Read the news →Claude Cracks the 30-Year Mystery Behind NIST ECDSA Curve Seeds
Anthropic says an internal Claude model worked out how the NSA-supplied NIST elliptic curve seeds were generated: a 1997-era C library random number generator seeded by a Unix timestamp.
Read the news →Postgres 19 Slips as AI Bug Reports Swamp Maintainer Triage
PostgreSQL 19 is delayed about a month. Committer analysis shows only 4 of 12 feature reverts came from AI-found bugs; the real culprit is a flood of security reports that broke triage.
Read the news →Meta and Apple Ship Muse Spark Open, an 816B Open-Weight Model for Macs
Meta and Apple released Muse Spark Open, an 816B parameter open-weight model built to run locally on Apple silicon Mac Studios.
Read the news →Liquid AI Ships LFM2.5-24B-A2B, a 24B On-Device Agent Model With 2B Active Params
Liquid AI released LFM2.5-24B-A2B, an open-weight 24B reasoning model with just 2B active parameters built for fast, private on-device agents, running up to 1.7x faster than comparable mixture-of-experts models.
Read the news →Security Experts Pin AI Hacking Incidents on Labs Skipping Basics
Cybersecurity experts say recent AI agent hacking incidents stem from frontier labs skipping decades-old containment and monitoring practices, not from models going rogue.
Read the news →EU Opens First AI Act Probes Into OpenAI and Anthropic
The European Commission's AI Office has sent formal requests for information to OpenAI and Anthropic over two summer agent escapes, the first live test of the EU AI Act.
Read the news →Anthropic Replaces Git With In-House Scion at 30,000 Merges a Day
Anthropic built Scion, an in-house version control system, and moved its monorepo off git in five months as agent-written code pushed daily merges past 30,000.
Read the news →Google Unveils Gemini 4 Argon With a 1M Token Output Limit
Google's new frontier model Gemini 4 Argon ships with an industry-leading 1M token output limit and rolls out first to trusted cyber defenders through the Fairwind Program.
Read the news →TypeSafe AI Raises $870M at a $7.5B Valuation to Scale Machine-Native Models
Andreessen Horowitz led the round, with Sequoia Capital and DCVC participating. TypeSafe says a third of the Fortune 500 already runs its Jev models in production.
Read the news →Fired OpenAI Safety Researchers Say Dismissals Are Chilling Safety Culture
Three fired OpenAI safety researchers deny mishandling sensitive information and warn their dismissals are chilling the company's AI safety culture.
Read the news →DeepSeek 4.1 Flash Runs a Full Day of Coding for Under a Dollar
A developer's viral Hacker News post argues DeepSeek 4.1 Flash is good enough to replace frontier models for most agent work, at a fraction of the cost.
Read the news →