This AI startup will give you up to $10 million if its AI tool fails to be productive

As companies rethink the use of AI tools due to rising costs, one AI startup, Cognition, has come out with "AI Productivity Guarantee." Cognition claims that if its AI agent, Devin, fails to deliver you value, the startup will make it up to you by giving up to $10 million (roughly Rs 95.7 crore).

In Short

• AI startup Cognition will pay you if its AI tool is not productive
• Every AI session to be evaluated and compared to hours a human would take
• You can get up to $10 million in credits

If you have browsed through the news about AI lately, you may notice a new issue. Companies are getting cautious when it comes to how much AI they use, largely due to rising costs. That is, some companies are worried that it may not be worth spending millions of dollars on AI, and are starting to put caps on how much AI their employees use – such as Uber and Walmart. To address this concern, the AI startup Cognition has come with a new idea – AI Productivity Guarantee. The startup claims that if its AI agent Devin does not deliver you value, you will receive up to $10 million (roughly Rs 95.7 crore).

Cognition announced this new guarantee on X, stating that “AI should earn its keep.” The company added, “If Devin delivers less engineering value than you’re paying for, Cognition will fund your usage until it does, up to $10 million.”

Devin is Cognition’s agentic AI that is aimed at being your virtual software engineer. You can delegate tasks to Devin, which then completes everything from start to finish in the background.

The startup claimed that now AI companies should no longer focus on tokenmaxxing, but deliver value. The post read, “It’s time for the AI industry to stop maximising tokens and start maximising productive output.”

Do note that companies like Amazon have started to crack down on how employees use AI tools after reports emerged that some were using AI for any task just to show higher token usage.

But how does this work?

On the surface, this may sound like a good idea. For companies who are starting to worry about high costs it can be positive to hear that they can get back millions if the AI tool fails to be productive. But measuring if the AI is productive may not be easy. And to do this, Cognition claims to have come up with a new system. This system does not focus on tokens, and instead measures productivity in hours.

In a blog post by Cognition CEO Scott Wu wrote that the company had built an AI estimator to measure the productive engineering output Devin provides to enterprise customers.

The system reviews each completed Devin session and asks two questions – would a human engineer have actually found this useful? And if yes, how long would that engineer have taken to do the same thing themselves? Cognition said it measures output in hours rather than lines of code because a small code change can still take hours of investigation, while some useful tasks may produce no code at all.

The company said the guarantee works by converting estimated engineering hours into a dollar value using a standard global rate. This is then compared with a customer's actual consumption near the end of an annual contract.

If the value falls short – that is, a customer has paid more than what this system shows the AI has done – Cognition will give up to $10 million to compensate for the gap in value provided. But this $10 million will be issued in credits, and not in cash. The credits can then be used by the customer in the future.

Though keep in mind that Cognition is evaluating the productivity from its own AI tools in this system, and then deciding the value.

How did Cognition come up with this?

Cognition states that it validated the system using human time estimates collected from enterprise users, covering 258 sessions from 126 users across a range of deployments.

The company added that individual estimates can be noisy, with errors of two to three times in either direction. However, these errors are expected to average out across many sessions.

This AI productivity programme is meant for Cognition’s entreprise customers that are deploying Devin Cloud at “meaningful scale” and meeting technical and engagement requirements, while existing customers can also enrol if they qualify.

Cognition also said the estimator is not a measure of full return on investment. It described the system as a baseline for productive output, saying ROI depends on broader business context and that hours saved do not by themselves capture business value or quality. The company said it plans to keep refining the system and publishing what it learns.

The announcement comes as investors and companies scrutinise AI costs more closely. Reports in recent weeks have said Microsoft has asked engineers to move away from Claude Code to an internal AI tool by June 30, while Uber executives have said the company exhausted its annual AI budget in five months.

OpenAI CEO Sam Altman recently stated that high usage costs for AI tools had become a “huge issue” for companies.

- Ends