Your Agent Isn't Dumb — It's Blind: agent-browser Takes Claude Code from 7 to 19

Most agent failures are not reasoning failures — they are fetch failures. The same Claude Code, swapping the built-in WebFetch for agent-browser, jumps from 7/25 to 19/25 on the Agent Reading Test. Same model, same prompt. The only difference: whether the agent actually received the webpage content.