Researchers found embedded browsers in Cursor, Copilot, Codex, VS Code, and Claude Desktop expose developer credentials. Two tools remain unpatched.

A community-built Laravel SDK for Tembo lets PHP apps start Claude Code, Codex, and Cursor agents, watch for completion events, and stream live output via WebSocket.

In September 2026, attackers took over a live AI coding session, slipped in a poisoned PyPI package, and used stolen GitHub tokens to self-replicate across 100 repos.

Spec-driven development gives AI coding agents written requirements to check against. Here is how OpenSpec and Spec Kit structure that workflow differently.

Google's Android Studio Rabbit 2 is now stable with support for 40-plus third-party AI coding agents via ACP, plus LeakCanary integration, XR emulators, and AI change review.

A developer learned the hard way that attaching a guardrail hook to write_file means nothing if the agent picks apply_diff instead. Here is the gap.

Researchers analyzed 51,412 refactoring instances across 12,256 pull requests to measure how AI agents like Claude Code and Cursor change code quality. The results are modest but statistically real.

Chidi Nwaogu shares engineering lessons from shipping an offline-first AI coding tutor on constrained Android hardware: storage limits, cold start, quota management, and failure-path design.

PostSharp 2027.0 ships an MCP plugin for Claude Code and Codex, CLI introspection commands, full build message docs, and an aspect test framework for .NET.

Developer Ashley vibecoded Lyrx, a native MacOS and iOS lyrics app, using GPT-4o Astra in Codex with no architectural guidance. Three prompts, open source, and a case for GPT over Claude.

GitHub announced Copilot Fusion, parallel agent tracking, and a free enterprise licensing tier at GitHub Universe 2026. Microsoft added 4M CPU cores to back it up.

OpenShell sandboxes the agent, Prempti judges every tool call, and Falco watches syscalls from outside. A Sysdig demo shows what each layer catches and misses.