NewAgents Just Need APIs — Benchmarking three ways to give AI agents web access: read the writeup →
Research · agent-data blog · issue 01

Research from agent-data.

Benchmarks, experiments, and evidence. What we measured, how we measured it, and what it means for agentic systems in production.

Empirical work from the agent-data team — benchmarks, ergonomics studies, and side-by-side measurements that compare how AI agents actually perform when they are given different ways to access the web. Each Research post ships methodology, numbers, and artifacts so you can replicate or push back. When the data is here, design choices stop being opinions.

FeaturedResearch

Agents Just Need APIs

Structured APIs, web search + extraction, and browser automation, measured side by side on a flight-search task. Structured APIs won every dimension: only modality to succeed, lowest cost, fewest turns.

9 min read · Read →
Subscribe

New pieces in your inbox, roughly weekly.

Short dispatches on building with agents. No launches, no recaps — real engineering. Unsubscribe any time.