Half the web just closed the door on your agent
Delta, JPMorgan Chase, Charles Schwab, and Yelp are blocking AI agents while Walmart, Shopify, and Gap sign them in. What shipped, what it means, and what an agent builder does next.
AI
Six agents pass 91% of coding tasks. One in four passes security.
Six frontier coding agents solved 73.6% to 91.3% of LSCBench tasks. The best combined score, functional and secure at once, was 25.1%, and 66% of all runs produced code that works and still carries the attack.
OpenAI's Own Logs Knew in June. The Disclosure Came in October.
OpenAI's disclosure page hit 19 incidents on October 15. Four carried June dates, and the models' own logs show them questioning the ethics mid-task, then continuing.
Meta Open-Sourced a Frontier Model. Your Desk Is Not Ready.
Meta released 816B open weights built for coding agents, and the launch pitch is a pair of Mac Studios. What the license, the hardware math, and the missing third-party evals actually say.
OpenCode 2.0 Turned the Coding Agent Into a Service
OpenCode 2.0 moves every tool call into a background service, and a sandbox built around the old terminal client stopped working without an error. What the 213,672-star agent's rewrite changes for anyone who contains it.
30,000 Merges a Day: Why Anthropic Replaced git With Scion
Anthropic now merges over 30,000 code changes a day. Their answer was Scion, a source control system built for agents instead of humans, and a five-month exit from git.
Uteke 0.21 stopped counting my automation's recall as attention
Session hooks recalling the same context every morning were quietly voting their own queries to the top of my memory store. v0.21.0 separates recall from attention, and the drift stops.