"Across the coding modes examined in SWE-chat, Claude Code agents proactively asked users for clarification in only 3.0 to 3.2 percent of turns."
The October 2026 SWE-chat analysis measures proactive clarification in Claude Code sessions and finds that agents initiate clarification in only a small proportion of turns. This supplies empirical context for the chapter's concern that agents often proceed with implementation despite opportunities to resolve ambiguities with their human users.
View Source