Payment for results could redefine the economics of AI businesses, says Bret Taylor

How does payment for results work?

In traditional software models, customers pay for access to a tool, typically through subscriptions or usage-based fees. However, AI agents operate on a different logic.

An agent is a system capable of performing tasks with a degree of autonomy, such as serving customers, retrieving information, and completing process steps.

Instead of charging for processing power, companies can set prices based on each completed outcome.

Sierra, a company co-founded by Taylor, already uses this model. According to Taylor, clients pay for tasks completed by agents, not for tokens.

In customer service, for example, charges may be tied to resolving a case without human intervention. In other sectors, outcomes could include loan origination or procedure pre-authorization.

Sierra calls its model outcome-based pricing. In a March 2026 interview, Taylor explained that charges apply only when an agent autonomously solves a problem. If human intervention is required, there is no charge for the service.

Why tokens may lose importance

Taylor’s proposed change addresses a key concern for companies investing in AI: how much value does the technology actually generate?

Different systems may use varying amounts of tokens to perform the same task. Therefore, focusing solely on processing volume doesn’t necessarily indicate which solution is more efficient.

In an interview with CNBC, Taylor noted that more advanced models can be more efficient in complex tasks, even if their cost per token is higher. He believes comparisons should consider the resources needed to complete a task.

This shifts the question companies ask. Instead of asking how much it costs to use a model, organizations can evaluate how much it costs to achieve a specific outcome.

Impact on professionals and companies

This change could also transform how companies evaluate AI projects.

Currently, teams may hire tools to summarize documents, answer emails, or analyze data, tracking metrics like user numbers, processing volume, or hours saved.

In a results-based model, evaluation focuses more on the final task. For example, a customer service system could be assessed by the number of cases resolved, rather than the number of interactions made.

For professionals, this means that knowing the tool won’t be enough. They will need to understand the process it performs, what results can be measured, and when human intervention remains necessary.

This logic also affects software providers. If payment is tied directly to outcomes, some risk shifts from the customer to the technology provider.

What changes in AI business

Taylor also argues that companies’ own data will be crucial in this new landscape. According to him, customer and business relationship information can become an asset that differentiates a company, even as AI models become more widely available.

The debate about outcome-based pricing isn’t limited to OpenAI. In March 2026, Taylor said this is Sierra’s model and compared the shift to the evolution of advertising, which moved from impressions to metrics tied to clicks and actions.

For companies, the challenge becomes precisely defining what counts as a result. A completed task may be measurable, but it still needs criteria to assess quality, customer satisfaction, and the need for corrections.

Taylor’s prediction suggests a potential shift in software economics: the focus moves from mere access to technology to the value it delivers. For professionals and companies, this makes the ability to measure results an increasingly critical part of adopting artificial intelligence.