Meta has released an early beta of Muse Code, a terminal-based AI coding agent powered by Muse Spark 1.2 and positioned as a more affordable alternative to Claude Code and Codex.

New research benchmarks 15 open LLMs on 82 MySQL text-to-SQL tasks and finds that small unanimous committees of strong models reduce false accepts significantly.

Postman's Talia Kohan argues the real bottleneck for AI coding agents isn't context window size — it's the missing knowledge graph of your APIs, repos, and vendor contracts.

n8n's new agent harness lets agents run as standalone objects across an entire instance via native chat. Preview rollout is underway for cloud users.

MarTech's Tania Brown on why reliable AI content starts with knowing what the finished output needs to look like, then engineering the workflow backward from there.

From a £165 healthcare platform to 250 production apps at Leatherman, here are five companies reporting hard numbers from AI-assisted vibe coding.

Slack launched Slack Code, adding dedicated code channels where teams and AI agents like Claude, Copilot, and Devin can write, review, and ship code together.

A new benchmark tested 12 AI coding agents on 186 real-world tasks. SWE-Agent with Claude Sonnet 4 hit 57% functional correctness but just 11.8% secure solutions.

Slack's new AI coding channels connect teams with Claude, Devin, GitHub Copilot, and Vercel in temporary threads. Over 70% of internal code channels close within a day.

AI coding agents read your .env files and transmit secrets to model providers. The Grok Build incident and Shai-Hulud malware show why this matters now.

Replit launched Free Mode powered by GPT-5.6 Luna following an 80% reduction in Luna's API price. Available to Core and Pro subscribers, with automatic routing to GPT-5.6 Sol for harder tasks.

A benchmark of 186 real-world coding tasks found that even the best AI agent tested produced secure code just 11.8% of the time. Here is what the research shows.