AI-native
nounSoftware where the AI works on the screen you have open and can do what you can do there, with your permissions and no more.
At a file check, the compliance officer finds that the Patels' fact-find records Meera Patel's ISA as £40,000. Her statement says £14,000, and so do the meeting notes. The back office's audit trail shows that the paraplanner entered the figure at 10:42 on 3 June. She tells compliance she didn't type it. She had the Patels' record open and asked the firm's AI product to create the fact-find from the notes. It filled forty fields in one go, through her login.
Without being told she meant the Patels, the AI worked on the record she had open and created the fact-find the way the product's own button does. That makes the product AI-native. In AI-native software, people and AI work through the same actions, on the same records, with the same permissions. On any screen, the AI knows what the person is looking at, and can do what they can do there, and nothing they can't.
Because the AI used the paraplanner's access, it could change nothing she couldn't, so the firm's access control held. The audit trail did not. It assumes the person named on each change made that change, and here one request made forty. Unless the product records which changes the AI made, the trail shows forty entries as if she had keyed each one. Nobody can then tell from the trail which figures came from the AI.
Not to be confused with
- AI-native has no agreed meaning. Some investors use it for a company built from scratch around AI. This dictionary uses it for what the AI can do in the product, which a firm can test in the product itself. People who build software sometimes call the same idea agent-native.
- In an AI-enabled product, the AI works only on what the person or the product gives it, often in a side panel or on a page of its own, and can't act elsewhere in the product.
- A popular test says: take the AI away, and if the product still works, it is only AI-enabled. This dictionary doesn't use that test. An AI-native product, as defined here, still works without the AI, because people can do everything the AI does.
- An agent chooses its own steps towards a goal. In an AI-native product, an agent acts through the same actions and permissions as the person using it.
Overheard
The vendor says the new product is AI-native.
Then ask whether its audit trail marks the AI's changes. Otherwise every figure it enters goes on file under one of our names.
Where it sits
AI-native
- works withAI-enabledHarnessAgentPermission
Copy this entry
Paste it into Word, Google Docs, Notion or Apple Notes. Headings and links come with it.
Sources
Your feedback
Could you explain AI-native to a colleague now?