A Pinterest staff engineer says the largest benefit from coding agents came from turning personal techniques into defaults the whole team could use. Jordan Cutler presented internal data showing merged pull requests among the top 10% of engineers rose 68% from early 2025 to late 2026, while the bottom half improved 14%.
Cutler first recorded instructions he kept repeating to agents and teammates, such as keeping pull requests small and testing from a user’s perspective. Sharing those rules helped new engineers avoid the same dead ends. The team then moved beyond individual sessions by scheduling agents to split large code migrations into pull requests for the appropriate teams.
One class-component migration ran three to five times faster than manual work, according to the presentation. The system also failed visibly: a looping job opened 16 pull requests in two minutes and pushed continuous-integration waits from under 30 seconds to 14 minutes. The response was to add a small guardrail rather than abandon automation.
Cutler says his team now ships roughly four times as much per engineer as comparable web teams using the same AI tools. The figures are internal and not an independent controlled study, but they highlight that shared process design may matter more than simply giving everyone the same assistant.