Vercel's evaluation found that embedding a compressed 8KB docs index directly in AGENTS.md outperformed the skills standard for teaching AI agents about Next.js 16 APIs, achieving 100% pass rate vs 79% with skills. The research shows that accessible, version-matched documentation embedded in the project beats explicit skill tooling for agent performance.
Research
AGENTS.md outperforms skills in our agent evals
Vercel's evals show that embedding a compressed 8KB docs index directly in AGENTS.md achieves 100% accuracy on Next.js API tasks versus 79% with explicit skills tooling.
Monday, April 6, 2026 12:00 PM UTC2 MIN READSOURCE: Vercel BlogBY sys://pipeline
Tags
research
/// RELATED
Products1d ago
Amazon’s trying to turn its massive shipping operation into another AWS
Amazon monetizes its private logistics network by offering supply chain services (freight, fulfillment, parcel shipping) to third-party businesses, directly competing with FedEx, UPS, and DHL with major customers like P&G and 3M already signed on.
WarApr 21
Nation-states want to cause harm, not just steal cash - stop handing your cyber defenses to the cheapest contractor
UK National Cyber Security Centre CEO warns that China now represents a peer-level competitor in cyberspace with sophisticated state-sponsored attacks, citing an average of four nationally significant incidents per we...