A single daily edition of announcements from the AI coding tools you use, read from a curated source list, summarised, and linked back to the original. No infinite feed, no account, no advertising.
Cursor loses direct access to OpenAI's models on 12 November
OpenAI told SpaceX it will wind down the contract supplying OpenAI models to Cursor, proposing a shutoff on 12 November 2026, and Cursor will also be blocked from future models including Astra. OpenAI said it lacks confidence that SpaceX will use OpenAI technology within its terms of service, citing X's contract breach after Musk acquired it and xAI's admitted violations. Co-founder Michael Truell said OpenAI models account for roughly 5 percent of Cursor traffic and that the company is working to resolve the dispute.
GLM-5.3's weights landed on Hugging Face, taking the coding model onto hardware you control
Z.ai's GLM-5.3 is published under the GLM-5.3 licence as a 753-billion-parameter mixture-of-experts model with dynamic sparse activation. The card lists 88.2 on Terminal Bench 2.1 and 28.3 on Terminal Bench 3.0, 78.1 on FrontierSWE, 66.9 on DeepSWE and 84.5 on CyberGym, with a reasoning-effort parameter taking low, high or max. Transformers, vLLM, SGLang, TokenSpeed, KTransformers and Unsloth are all supported and Docker deployment is documented, but hardware requirements are not stated.
The gap between a bug being discussed and your server being probed is about ten minutes
Anil Madhavapeddy stood up a site after a traversal bug was discussed publicly and reports it was soon fielding probes for percent-encoded traversal sequences; he located the exploit with DeepSeek V4 Pro after Claude Fable declined. Nick Craig-Wood says rclone took about 20 security disclosures through GitHub in its first ten years and over 40 in the last month, at a 75% hit rate, while GitHub's CVE assignment has slipped from two or three days to three or four weeks. Willison's conclusion is that open-source embargo practice no longer fits the timeline.
Agent sessions started in another app now resume inside VS Code
VS Code 1.135 lists recent Copilot or Claude agent sessions created in other applications, so a session started elsewhere can be continued in the editor with a Copilot subscription, gated behind a chat.agentSessions.showExternal setting. Microsoft also introduced Agent Host, which connects the same agent session from multiple VS Code windows, runs harnesses over the Agent Host Protocol and integrates the Copilot SDK to align behaviour across Copilot CLI and GitHub's standalone app. An experimental rubber-duck command asks a complementary model for a second opinion on the agent's work.
