Cognition says it is backing a claim about its coding agent Devin with up to $10 million per customer, framing the money as part of a private metric designed to measure engineering output.

The company shared the message in a social post from Ryan Bai, which linked to an article titled, "$10,000,000 on the line: how we measure Devin’s engineering output." In the post, Cognition said it is putting up a large incentive behind a single assertion: that Devin produces more engineering output than customers pay for.

The company said the measurement system was built to support that claim and that it has been validated using independent data. The post did not provide the full methodology, the identities of the customers involved, or a detailed explanation of how the output metric is calculated.

Cognition’s announcement is notable because it attempts to quantify the value of an AI coding agent in business terms rather than relying only on benchmarks or general performance claims. In the software industry, vendors often argue that AI tools improve productivity, but those claims are hard to compare across teams and projects. By linking Devin’s performance to a financial commitment, Cognition is signaling confidence that the agent can generate measurable engineering value.

The company did not say in the post how the $10 million figure would be distributed, whether it represents a guarantee, a performance pool, or another contractual arrangement. It also did not disclose whether the incentive is available to all customers or only specific enterprise accounts.

Devin has been one of the most closely watched AI coding systems in the market as companies race to build agents that can assist with software development, automate routine tasks, and handle parts of the engineering workflow. Claims about these systems often focus on speed, code quality, and developer time saved, but direct evidence of economic impact has been harder to establish.

Cognition’s framing suggests it wants to move the discussion from abstract capability toward measurable results. By tying the product to a customer-facing financial claim, the company is effectively inviting scrutiny over whether the agent can reliably increase output in a way that justifies its cost.

The post was short and directed readers to the company’s article for more detail. Based on the public message alone, Cognition has not yet opened its measurement system to broader review, but it has indicated that independent data played a role in validating the approach.

For now, the announcement places Devin in a different category from many AI tools marketed with general productivity language. Instead, Cognition is attaching a concrete dollar figure to the question of whether the system can outperform what customers pay. That makes the metric itself as important as the product claim behind it.