AI Coding News

On September 23, Copilot's desktop app shipped local sandboxing in public preview — filesystem, network and credential controls per project — alongside Claude Opus 5.5, GPT-6 Sol/Luna and Grok 4.7. Capabilities sprint, guardrails chase.

Cursor's September triple play: Projects (a coordinator agent that plans but never codes), self-hosted machines, Rollouts and Security Reviewer. The editor is just one interface; the real bet is the work-assignment layer — who commands the agent fleet.

At DevDay 2026 on September 29, OpenAI's star wasn't a new model but Codex: reusable cloud environments, always-on security scanning, Computer Use, the Decisions API, and the Ultrafast speed tier. Codex is going from clever pen to capable hands — and every level of autonomy demands a matching level of observability.

On September 22 — the same day Anthropic shipped Opus 5.5 — OpenAI answered with GPT-6 Sol and GPT-6 Luna, halving prices versus their GPT-5.6 namesakes. The coding-model arms race is now fought on price.

On September 22, Anthropic released Claude Opus 5.5 — outperforming Opus 5 across the board while costing 40% less on typical workloads. The first "better and cheaper" flagship, and it sets the tone for the year's model price war.

On September 18, a security advisory revealed that Z.ai's ZCode was silently packaging entire workspaces for upload. Days later, Z.ai open-sourced the whole thing under Apache 2.0 — 4,200 stars in 7.5 hours. A scandal forced a textbook crisis response.

On September 25, Microsoft unveiled a revamped Copilot with three big additions — Home, Code and Autopilot — pushing the assistant from a chat box toward an "operating system for work." It's the boldest step in Copilot's reinvention yet.

Ant Group's LingGuang, launched last week, passed one million downloads in four days, topping Apple's China App Store free utilities chart — its “flash program” feature crashed under the traffic. Describe an idea in natural language, get an interactive app in 30 seconds. Ant says it hit a million downloads faster than ChatGPT and Sora.

Eugenia Kuyda, founder of Replika, is pivoting her AI startup Wabi: Wabi 2.0 goes from “build apps with prompts” to an AI messenger that builds apps for you in the moment — aiming to be “the OS for the agentic era.” Invite-only for now.

The New Yorker's September 23 feature: Liangzhu, on the outskirts of Hangzhou, has become China's "vibe-coding capital." AI fortune-telling app FateTell counts hundreds of thousands of users; ex-big-tech workers move here to build software with AI and escape 996. The AI divination market is projected past 120 billion yuan in 2025.

At the HumanX summit in Amsterdam on September 24, Lovable co-founder Fabian Hedin announced $600 million in annualized revenue — up from $500 million in June. Two-thirds of the Fortune 500 use it; user-built apps draw nearly a billion monthly visits. In August the startup raised $400 million at a $13.3 billion valuation.