Cognition, the developer of the AI coding agent Devin, has launched an internal system to evaluate engineering efficiency. The tool calculates a score by dividing the estimated human-equivalent hours of a session by the total cost in tokens and spend. Engineers who meet specific productivity benchmarks receive dynamic increases in their spending limits.

The company validated its estimation model using feedback from 126 users across 258 sessions, achieving a log correlation of 0.74 on held-out data. This system, which went live in mid-2026, aims to shift focus from high token consumption to tangible output. Reports detailing the methodology were released on October 15 and 16, 2026.

Alongside the grading system, Cognition introduced Devin Coach, an assistant designed to help users optimize prompts and break down complex tasks. This initiative complements the company's AI Productivity Guarantee, announced in June 2026, which provides up to $10 million in credits if customers do not receive value equivalent to their spending.