AI-native

On this page

Look at the delivery method

Consider two constructed document services. Both use the same model and both have human staff.

In the first, a consultant occasionally asks AI to summarise a document, then follows the existing manual process. In the second, incoming documents enter an AI-supported extraction and checking process, with unresolved cases routed to an appropriate reviewer. The offer, work allocation and handling of exceptions are organised around that delivery method.

The second illustrates what we mean by AI-native. The distinction is about the operating design, not a percentage of employees replaced or the presence of a chatbot.

An emerging label, not a certification

Rex describes itself as an AI-native service for order-to-cash work, with people handling approvals and exceptions. That is a company’s account of its approach, not proof of a universal operating model. The a16z essay cited below develops the investment thesis that software can perform work previously bought as labour.

Our definition draws on that usage. It does not establish a standard, guarantee margins or imply that every service should use AI.

Ask what the customer actually receives

For the hypothetical document service, the promise might be checked structured records with unresolved items identified. Assess whether customers value that output and whether checking, rework and support leave a viable service.

The service could still sell through ordinary human relationships. Making it accessible to software agents is a separate agent-native design question.