Cognition's AI Efficiency Grading System Cracks Down on Tokenmaxxing

Precedent & Market Context

Cognition, the startup behind the autonomous coding agent Devin, has a history of innovation in the AI space. In June 2026, the company introduced its AI Productivity Guarantee, which offers up to $10 million in credits if customers do not get value equivalent to what they spend. This promise reflects a broader rethink across the industry about what AI productivity metrics should actually reward.

Main Event

Cognition's new system grades engineers on their efficiency, using a combination of human-equivalent hours and token consumption. The system was publicly detailed in reports published on October 15 and 16, 2026. Engineers who hit defined productivity quotas see their spending limits rise dynamically, flipping the logic of tokenmaxxing on its head.

Analysis & Market Impact

The efficiency grading system is a significant departure from the traditional tokenmaxxing approach, where heavy token use was seen as a signal of a power user. Cognition's pivot reflects a growing recognition that this approach can lead to inefficient use of resources. The system's use of AI-powered efficiency grading also raises questions about the potential for gaming and misfires.

Implications & Outlook

Cognition is betting that customers will prefer to pay for results rather than volume. The company has put up to $10 million in credits behind this bet. However, the success of the efficiency grading system will depend on its ability to accurately measure efficiency and provide clear feedback to engineers. If successful, the system could lead to a shift in how companies approach AI productivity metrics.