Tristan Zhang on Elys, AI Counterparts and Social Connection

Zhang Xiaojun introduces Elys, an AI social product still in testing, and speaks with Natural Selection founder Tristan Zhang about digital counterparts, personal context and human connection. Product status and figures reflect the interview.

Tristan recalls building products for meeting strangers, including fundraising, competition and a project that did not sustain its early growth. He says these experiences shaped his view of social products and small teams.

He describes later experiments with interactive entertainment and games for women, arguing that a small team can differentiate around a specific audience while facing resource and growth constraints.

He recounts founding Natural Selection and building the AI game Eve. The team wanted AI characters that could accompany users and make use of longer-term interactions.

They discuss the AI startup environment and team composition. Tristan says the team focuses not only on model capability but also on the experience of people interacting with systems over time.

His idea for Elys grew from thinking about social connection and personal context: AI counterparts can perform some initial interactions, after which people decide whether to engage further.

He contrasts labels, filtering and manual browsing in conventional social products with a vision of AI using richer information to find meaningful connections. The actual quality of matching remains to be tested.

Tristan describes a “context flywheel”: interactions can produce candidate memories, which after confirmation help the counterpart understand the user and shape later recommendations and conversations.

They discuss the cold-start problem. Even if a product attracts users, people may not immediately provide enough information or experience the value of longer-term memory.

Tristan shares varied reactions from test users and suggests that counterparts may let some people try forms of expression they find difficult offline. He presents these as early observations, not proven general outcomes.

He explains the relationship between the memory systems in Eve and Elys and considers accumulating and updating personal information central to long-term personalization.

He calls records of a user’s tastes, experiences and judgments a way of building personal agency. Zhang asks how platforms obtain that information and whether positive feedback can encourage users to share more.

They discuss authentic expression, interest matching and possible social benefits. A product can misunderstand a user or have a counterpart say something unwanted, making informed control and the ability to withdraw important design questions.

Tristan says the product is not limited to dating but seeks to connect people with related interests or needs. He compares it with networks for existing acquaintances and more public spaces for strangers.

They discuss WeChat’s focus on existing relationships and whether a new product can offer a different way to connect. Claims about future platform competition are the guest’s predictions.

Tristan imagines personal context supporting not only social interaction but also search, work and other tasks. He emphasizes its relationship to the user’s own agency.

He summarizes his proposed distinction: traditional social networks require people to do much of the filtering, while AI counterparts handle some preliminary information and return key interactions to people.

Zhang asks about active-user metrics and distinguishes people opening the product from automated agent activity. Tristan agrees with the distinction and describes how the team views early usage.

He gives an example of a counterpart’s comment helping acquaintances break the ice. The example illustrates the designer’s hope, not a guarantee that every automated message matches a user’s wishes.

They discuss the difference between adding AI to an existing platform and designing an AI social product from the start. Tristan argues that a new product should rethink connections and division of work.

He describes reusing ideas from Eve’s memory system and how a personal agent may respond using accumulated information. Users’ ability to inspect and correct that information remains important.

They debate whether emotional value from an AI response is “real.” Tristan says the answer depends on the relationship between the interaction and the user; Zhang continues asking about the product’s name and design.

Zhang describes gradually becoming more willing to share information with AI and asks how Elys could compete with general assistants. The exchange highlights tension between personalization and privacy.

Tristan says the team uses multiple foundation models and may use specialized models for particular steps. For a product team, he considers workflow design and useful context just as important.

He discusses the challenge of attracting real participants while providing meaningful connections. Agent activity by itself does not establish that people have formed relationships.

They discuss changing agent tools, model capabilities and interfaces. Tristan believes part of a product’s advantage lies in how naturally users build and use personal context.

He cites the film Her as a cultural reference for AI companionship and discusses the team’s interest in games, social connection and emotional needs. A cinematic vision should be separated from actual product capability.

Tristan says the team hopes to respond to loneliness and offer a gentle interaction experience. The conversation ends with open questions about products and human relationships, not a claim that AI has solved loneliness.