What happens when you give AI coding agents a lab full of robotic arms, some compute resources, and a “generous token budget” for teaching the robots various tasks? The agents can apparently figure out a training regimen that teaches the robots to successfully cut zip ties and even insert GPUs into thin sockets on motherboards.

The first proposed catalog of 'configuration smells' reveals widespread issues like context bloat, skill leakage, and conflicting instructions that can make coding agents less reliable and more expensive.

AI coding agent skills library claude-skills ships 345 free, MIT-licensed packages for Claude Code, Codex, Cursor, Gemini CLI, and 9 more tools, including 51 senior engineering roles, a hostile adversarial code reviewer,

Anthropic's tools are getting chummy with each other.

A new framework, Arbor, they claim, preserves hypotheses, experiments, and lessons learned across long-running research tasks, delivering 2.5x better performance than other models under the same budget.

Cohere's North Mini Code ranks 8th of 127 open-weight models on output speed — but generates 3x the output tokens of comparable models in independent testing.

AI coding agents boost code output by 180% but shipping rises only 30%, MIT finds. Why private data access beats benchmark scores as the real AI investment moat.

Researchers warn Agentjacking can abuse Sentry errors to make AI coding agents run malicious code on developer machines.

Don't know coding? No worries. You can still create your AI Agent. Let us see what we need to do to make a brand-new AI agent.