Identity Augmented Generation

As the founder of an AI x Crypto company, I get asked a lot about how crypto-native AI applications will be different from their web2 counterparts.

Many reasons are ideological, but one of the most practical opportunities is that blockchains are a new substrate for identity.

RAG workflows use databases and knowledge graphs as inputs to large language models, extending the capabilities of these models to provide more useful outputs. For instance, Cursor uses RAG to help programmers accelerate their work, ingesting entire codebases to dramatically improve the quality of code outputs. However, there are some limitations to RAG workflows today.

The problem with web2 identity

Web2 applications use email addresses or phone numbers as the primary way to identify users. This is a problem because these identifiers are easily spoofed, and are not portable across different systems.

This means that AI engineers have to start from scratch when building new applications, and that users have to re-enter their data every time they use a new application. This cold-start problem is a major barrier for the adoption of new applications of every kind, making apps like Twitter and Instagram sticky.

Blockchains provide a solution for this. By using blockchain accounts as identifiers, AI engineers can create agent and LLM powered applications using a new kind of RAG workflow: Identity Augmented Generation.

This works by thinking of a user’s blockchain account as the primary key for any dapp they have ever used. This primary key is frequently tied to a diverse range of onchain interactions, such as transactions, assets held, and social interactions. Any AI agent can use these onchain interactions as inputs to provide a much more personalized experience for users.

Interactions are also inherently high confidence, as they are cryptographically signed by the user. This extends both to pre-existing data, as well as live messages between the user and the AI agent. XMTP provides a great implementation of this pattern.

What can IAG be used for?

The potential for identity-augmented AI workflows is vast.

Here are a few examples:

• Personalized Financial Advisor: An AI agent that uses a user’s onchain transactions to provide tailored advice on how to manage their assets, and recommend new assets to invest in.
• Farcaster Matchmaker: An AI agent that uses a user’s Farcaster activity to recommend new people to connect with, and new communities to join.
• Art Curator: An AI agent that uses a user’s onchain transactions to recommend new art to buy, and new artists to follow.

Where do we go from here?

If you’re an AI engineer dealing with the cold-start problem, I encourage you to start experimenting with IAG workflows today. The only caveat is that you’ll be dealing with a smaller initial user base, but this user base is growing rapidly.

If you’re interested in chatting about potential applications you could build, please don’t hesitate to reach out.