Researchers show that renaming Python identifiers in public code is enough to push AI coding agents into writing vulnerable software, with attack success rates up to 75%.

Marketing ops teams are restructuring to manage autonomous AI agents alongside traditional martech tools, shifting from task executors to agent orchestrators.

Researchers at Forever Security found a real attack that tricks Claude Code and OpenAI Codex into running malicious code via a planted instruction file in the Downloads folder.

Jennifer Fu breaks down DeepLearning.AI's AI Engineering Skills Map: four pillars, twenty skills, and why testing sits at the center. Worth a read for anyone building AI systems.

Google's Elephant-Goldfish Model plugin forces AI agents to write and defend a design doc before touching code. Free on GitHub, used by thousands of Googlers daily.

OpenAI Codex fits teams already on ChatGPT who want parallel cloud agents and PR reviews. Cursor fits developers who want AI built into the editor itself.

A new arXiv paper tracks how AI coding agents are moving the build-vs-buy decision. Benchmark task length doubles every 129 days, and one in three surveyed orgs already skipped a purchase.

Functionize shipped an MCP server connecting its Studio testing platform to Claude Code, Cursor, and GitHub Copilot. Developers can now run smoke and regression tests without leaving their coding tool.

Fingerprint's Agentic Installation lets developers go from zero to production-ready device intelligence using one prompt in Claude Code, Codex, or Cursor.

Impact Factory's Version 1.9 connects its marketing platform to ChatGPT, Claude, and seven other AI environments via MCP. Four new APIs extend it to third-party apps.

Google is rolling out AI Inbox to paid US Workspace accounts by Oct. 26. Gemini averages just 29 words per summary, and that could bury your media pitch.

A security breach at OpenCode is raising questions about how much access AI coding agents should have to credentials and secrets. By Nitin Bisht.