The State of Agentic AI

Anthropic’s landmark study, published February 18, 2026, provides the first empirical snapshot of how AI agents actually operate in the wild — analyzing nearly one million tool calls from their public API and 1.5 million Claude Code sessions.

The findings confirm, with hard data, what we’ve been tracking through our analysis of the agentic economy: we are in the early innings of a structural transformation, and the autonomy models can already handle more than what users grant them.

This analysis synthesizes the Anthropic data with the strategic frameworks we’ve developed — the barbelled distribution economy, the AI-in-human-loop paradigm, the agentic web architecture — to answer a single question: where exactly are we with agents, and what comes next?