What's this about?
From the same study: Codex web-searches in 94% of sessions (usually with `site:` operators), Cursor in about two-thirds, Claude Code in only ~30% — but when Claude Code does search, it browses 3× more pages. The three agents pick the same tool in just 42% of cases.
What exactly does 42% agreement mean?
There's no objective "right tool" the agents converge on — vendor choice is mostly an artifact of each harness's research style and defaults. The harness decides, again.
Why is this worth noting?
For teams: which agent you adopt quietly shapes your dependency tree. For vendors: three different marketing channels, one per agent. For the ecosystem: 42% is the number that makes "agent-influence" a real industry.