Cognition is taking an unusual route to win over enterprise customers that are becoming more cautious about AI spending. Instead of only promising more usage, the company is backing Devin’s productivity with a guarantee that can reach as much as $10 million if the software engineering agent fails to deliver enough value.
The move lands at a time when many companies are questioning whether rising AI costs are producing enough real output. That pressure has pushed Cognition to frame Devin as a tool that should be judged by results, not by how many tokens it consumes.
A guarantee tied to output, not usage
Devin is positioned by Cognition as an AI agent for software engineering work. Users can hand over tasks, and the system is designed to complete them end to end in the background.
In a post on X, Cognition said AI should generate value rather than simply encourage heavier consumption. The company described the approach as an AI Productivity Guarantee and argued that the industry should stop maximizing tokens and start maximizing productive output.
That message fits a broader shift in how enterprise buyers are looking at AI. Several companies are already applying stricter controls as they look for a clearer return on investment.
Why the pricing pressure matters
Uber and Walmart are among the companies said to have started limiting AI use among employees because the gains have not always matched the cost. Amazon is also said to have tightened employee use of AI tools after reports that some workers were using AI for nearly every task in an effort to drive higher token usage.
The wider market has also become more sensitive to the economics of AI operations. OpenAI CEO Sam Altman recently said that high AI usage costs have become a major problem for companies, while another report said Microsoft asked engineers to move from Claude Code to internal AI tools before June 30.
Uber executives have also said the company’s annual AI budget was exhausted in five months. Against that backdrop, Cognition’s guarantee looks designed to reassure customers that they are paying for results, not just activity.
How Cognition measures Devin’s value
Cognition does not base Devin’s value on token counts or the number of code lines produced. Instead, it uses estimated engineering hours generated through the system.
In a company blog, CEO Scott Wu said Cognition built an AI estimator to measure the productive engineering output Devin provides to enterprise customers. Each completed session is reviewed through two main questions.
The first asks whether a human engineer would consider the result useful. If the answer is yes, the second asks how long that same engineer would likely need to finish the task alone.
Cognition said it chose an hours-based approach because even a small code change can require hours of investigation. It also noted that some tasks can still be useful even when they do not produce any code at all.
Once the hours are estimated, Cognition converts that output into dollar value using standard global rates. That figure is then compared with the customer’s actual consumption near the end of the annual contract.
If the customer paid more than the productivity value Devin delivered, Cognition says it will cover the difference up to $10 million. The compensation comes as credits for future use, not as cash.
Who can qualify
The program is aimed at enterprise customers using Devin Cloud at what Cognition calls meaningful scale. Customers also need to meet technical and engagement requirements, and existing customers may join if they qualify.
Cognition said the system has already been validated using timing estimates from enterprise users. The validation covered 258 sessions from 126 users across different implementations.
The company acknowledged that individual estimates can be noisy and may be off by as much as two to three times in either direction. It argues that those errors should partly balance out across a larger sample of sessions.
Even so, Cognition said the estimator is not a full measure of return on investment. It described the metric as a baseline for productive output, since ROI still depends on broader business context and the number of hours saved does not automatically capture business value or result quality.
The company said it will continue refining the system and sharing what it learns. In a market now focused as much on proving value as on expanding AI use, Cognition is betting that a productivity guarantee will stand out.
Source: www.indiatoday.in
