Mar 1
Loading video player...
MCPs, skills, tools, rules — there are so many ways to give context to your coding agent. But which one actually works best for building NextJs 16 apps? The NextJs team ran real evals comparing skills vs agents.md, and the results might surprise you. Even with explicit instructions, skills only hit a 79% pass rate. agents.md with a progressive disclosure pattern outperformed them all. In this video, I break down: → What MCPs, skills, and agents.md actually are → Why skills sometimes get ignored by your coding agent → The retrieval-led reasoning trick that changes everything → How to set up agents.md with a compressed docs index matched to your NextJs version → The progressive disclosure pattern to keep your context window clean 🔗 Resources mentioned in this video: Next DevTools MCP: https://github.com/mcp/vercel/next-devtools-mcp Vercel skills repository: https://skills.sh agents.md blog post: https://vercel.com/blog/agents-md-outperforms-skills-in-our-agent-evals AI agent evaluations: https://nextjs.org/evals 💬 Have you found a workflow or pattern that works well for your AI-assisted development? Drop it in the comments — I'd love to hear what's working for you. #nextjs #nextjs16 #ai #codingagent #webdevelopment #agentsmd #mcp #claude #cursor