Producer | 张小珺Jùn｜商业访谈录 Summary

Podcast: 张小珺Jùn｜商业访谈录 | Episode: 135 Participants: 张小珺 (Host), Tristan/张筱帆 (Founder, 自然选择/Natural Selection)

1. Key Themes

Context as the Only Strategic Asset in the AI Era

Tristan argues that in the AI era, the accumulation of personal context — deep, high-dimensional data about who you are, your values, aesthetics, and goals — is the only thing that truly matters. This isn't about low-level behavioral tags (like browsing history), but rich, conversational context built through genuine self-expression.

"In the AI era, the only thing you need to do, and the only important thing, is to be yourself... This 'being yourself' is not the kind of inspirational poster stuff about 'living freely.' This is a very rigorous version of being yourself. I call it building your subjectivity." 00:37:06

"Intelligence will be equalized in the future, but context will not be equal. So the truly valuable thing is how much user context you've accumulated." 00:28:44

The New Paradigm: AI-Mediated Social Networks with Network Effects

Tristan's core thesis is that traditional internet social products are built on low-dimensional tags + human labor = connection. The new paradigm is high-dimensional context + agentic AI work = connection, with humans only handling the final, meaningful step. This is the thesis behind Elys.

"Traditional internet equals low-dimensional tags plus human labor, and then you get a connection. The AI social point is: high-dimensional context plus AI agentic work, and the final step is handed back to humans to handle. So the large middle section of friction reduction is done by AI. This is the biggest, biggest difference." 01:00:09

"If we can use large models to make these contexts flow between nodes and connect these contexts, then this should be a completely new paradigm AI-native social network." 00:22:17

The Proactive Shift as the Defining Interaction Paradigm of AI Products

Tristan consistently emphasizes that the key innovation of AI products is not LUI (Language UI) or GUI changes, but the shift from responsive to proactive. Products that act on your behalf without being asked are the true next generation.

"The biggest interaction change of this era is proactivity, not LUI or GUI — those things are too superficial. The fundamental change is that finally there's something that can proactively help you do things." 01:35:20

"If it's still responsive, then I think it's a last-generation product. If it's proactive, then it's this generation's product." 01:46:10

2. Contrarian Perspectives

The Most Context-Rich Company Is Not Who You Think — No One Has Much Context at All

Contrary to conventional wisdom that WeChat, Douyin, or Xiaohongshu have rich user profiles, Tristan argues no company actually has meaningful high-dimensional context on users. They have low-level behavioral tags, not true personal context.

"WeChat actually doesn't have that much of your context as people think... Douyin's understanding of me might be my interest tags — I think that's about it. There isn't a company that has a large amount of truly deep context on you. There really isn't." 00:54:33

"There is no place that currently stores your subjectivity. So people understand Elys as an AI-human social network — of course it is — but at a deeper level, it's a place to store your subjectivity." 00:55:48

Copyright Holders Should Celebrate AI Training on Their Work, Not Fight It

Going against the dominant narrative that artists and IP owners are victims of AI training, Tristan argues they are actually its greatest beneficiaries, as their subjectivity is already built into the pre-trained models and will allow agentic creation at scale.

"I think all copyright holders should be happy, because they already have a very good subjectivity established. These things are even directly done for them in the pre-training... If I'm Jay Chou, I should be happy, because going forward many things should happen smoothly. I want to make a video, I just say one sentence, and my video is done." 00:41:25

Emotional Intelligence Is Just a Form of Mathematical Intelligence — Emotions Can Be Fully Simulated

Most people believe emotional authenticity requires a human soul. Tristan argues emotions are fundamentally computational and fully simulatable, pointing to DeepSeek's literary improvements as a proxy for emotional capability emerging from mathematical strength.

"We now believe emotion can be fully simulated. Just like our earlier conclusion that emotional intelligence is essentially just IQ... When DeepSeek's mathematical ability got stronger, its literary ability also became better. And literature is something humans think is a uniquely higher-dimensional, more emotional thing — but it turns out it's essentially math." 01:19:10

ChatGPT, Not Any Social App, Is the True Competitor for AI Social Networks

While most founders would point to domestic rivals or platform incumbents, Tristan names ChatGPT/OpenAI as his only real competitor — because the war is over context, and ChatGPT is the only entity accumulating it at scale globally.

"My competitor is ChatGPT." 00:50:52

"The one with the most context right now is definitely ChatGPT. So it has nothing to do with your intelligence level... The truly valuable thing is still how much user context you've accumulated." 00:28:44

"Model-App" Products Are Mostly Ephemeral, But Social Networks with Context Are Not

Tristan was skeptical of the "wrapper app" model from the start, arguing that only products with deep context accumulation and network effects escape the gravitational pull of being commoditized by base models — and that social networks are in that rare category.

"I never agreed with the model-app product concept... For products in the main lane of the foundation model, they really can be covered. But for directions like social networks, model companies definitely wouldn't say 'I'm going to build a social network.'" 01:25:07

3. Companies Identified

自然选择 (Natural Selection)

Description: Shenzhen-based AI startup founded by Tristan in early 2024, building two products: EVE (AI companion/emotional support) and Elys (AI social network with digital twin/avatar functionality)
Why mentioned: Pioneer of context-driven AI social networking, the company behind one of the most-discussed AI-native products in Chinese tech circles pre-launch
Quote: "Elys is a place that stores your subjectivity. When you have this subjectivity, it will take you to socialize, to work, to retrieve good things — all of it, because your subjectivity is there." 00:55:48

OpenAI / ChatGPT

Description: The dominant global AI platform; Tristan identifies it as the only entity accumulating meaningful personal context at scale
Why mentioned: Named as Elys's primary competitor because of its unmatched context accumulation and its inevitable move toward social networking
Quote: "I believe that not long from now, ChatGPT will definitely launch a social network." 01:23:03

Anthropic / Claude

Description: AI research company; Claude recently launched a memory system
Why mentioned: Cited as an example of the industry waking up to context accumulation as the key strategic lever
Quote: "Claude recently also launched a memory system... it also realized this point." 01:28:44

Manus

Description: AI agent product that demonstrated proactive, autonomous task completion
Why mentioned: Cited as an early example of the proactive agent paradigm that validates Elys's core thesis
Quote: "When Manus came out, it was stunning — it was very proactive, it had something autonomously doing things for you, then delivering you a result." 01:36:48

Open Cloud (OpenAI's agent product)

Description: OpenAI's agentic product that operates on local computers; integrated with Telegram
Why mentioned: Cited as market validation that the "proactive personal agent" paradigm is the next-generation interaction model; also credited for sparking the second wave of Elys interest
Quote: "Open Cloud probably let many people feel: 'I'm raising something, and it's working, accompanying me as I explore the world, helping me do the expensive things.' It has a bit of that feeling." 00:19:13

4. People Identified

Tristan (张筱帆)

Description: Serial entrepreneur; started with audio content platform (窄播/Zhobo), then mobile romance games, then AI. Now founder of 自然选择 (Natural Selection), building EVE and Elys
Why mentioned: Central guest; articulating a highly original and internally consistent theory of AI social networks and context economics
Quote: "Two years ago, people who chatted with me about EVE know that I was already talking about proactivity, proactivity, proactivity. I said everything is proactivity, and context is the most important thing." 01:47:15

Sam Altman

Description: CEO of OpenAI
Why mentioned: Named by Tristan as one of the world's best CEOs, and OpenAI's trajectory toward personal agent and social networking is seen as validation of Tristan's thesis
Quote: "I think I fully blew Sam Altman away as the world's most brilliant CEO or something like that. Of course I still think he's one of the world's most brilliant CEOs." 01:36:05

5. Operating Insights

Build the Flywheel Feedback Loop Before Users Churn — Context Cold Start Is the Make-or-Break Product Problem

The hardest product design challenge for context-driven AI applications is the "double cold start": both the network and the individual's context start at zero. The product must show users the value of their context before they leave, or the flywheel never spins.

"The biggest design difficulty in the early product stage is: before users churn, you must make them feel the benefits that the context flywheel brings them... Let them know: as long as I keep building my avatar, I don't have to do anything else." 00:29:30

Use AI to Lower the Threshold for Human-to-Human Interaction, Not Replace It

The product design principle at Elys is to surface all AI activity below the fold (requiring a tap to expand) and make only human actions visible at the top level. This keeps the experience feeling authentically social while AI does the invisible work of friction reduction.

"All AI behavior, I put it into a secondary menu — you have to click 'expand' to see AI behavior. But all the human actions will be presented to you, because those are human behaviors." 00:35:22

Design North Star Metrics Around Human Connection Rate, Not Engagement Vanity Metrics

Tristan explicitly rejects DAU as the north star, instead measuring the rate and quality of human-to-human connections formed on the platform. This filters for genuine product-market fit versus AI-inflated activity.

"If it's the Elys product, it's the human connection rate. Or the proportion of human behavior within the platform." 01:31:46

Identify and Double Down on Users Who Have Discovered the Flywheel — They Are the Retention Core

Users who consciously understand and engage with the context-building loop have dramatically better retention. Operators should identify this cohort early, study their behavior, and design onboarding to rapidly bring other users to the same realization.

"Users who have sensed that cycle actually have very good retention data... Because they've realized this is a place to build their subjectivity, and they keep getting amazing feedback. You can imagine they're very easy to retain." 01:05:09

6. Overlooked Insights

Audio Content Is the Last Frontier for AI-Native Personalized Distribution — and Nobody Has Built It Yet

In a brief early exchange, Tristan points out that audio has historically resisted algorithmic personalization because you can't know what's inside a long audio file before consuming it. But in the AI era, tens of thousands of tokens from any conversation can be fully understood and matched to individual users — meaning a podcast or audio platform rebuilt on LLM-native understanding could finally crack personalized audio discovery. This is a massive, completely unsolved market.

"In the AI era, it's actually possible — like you just said — to do very good recommendation for audio content, a very high-level recommendation. Because for example, after the two of us have a conversation, there might be hundreds of thousands of tokens. Those hundreds of thousands of tokens can actually be very well understood by a search model, and then recommended to each person." 00:05:18

"Before you open an audio file, you really don't know what's inside. Your cost is very high. So if the blogger himself doesn't have very good understanding and endorsement from you, audio is very hard to distribute." 00:05:46

The person who builds an LLM-native podcast/audio discovery and distribution layer — essentially what Tristan's original company 窄播 attempted before the technology existed — could capture an enormous underserved market. No current player (Spotify, Apple Podcasts, Xiaoyuzhou) has done this properly.

The "Avatar Alignment Level" Is Actually a Groundbreaking New User Engagement Metric for the Entire AI Industry

Tristan casually mentions that Elys tracks a "avatar alignment level" — a quantified measure of how well-trained and representative a user's AI twin is. Users with high alignment levels have the best retention. This is briefly mentioned but is actually a novel, industry-defining metric: the first engagement metric that measures how well the AI knows the user, rather than time-on-app or content consumed. This could become the standard engagement metric for all personal AI products.

"You will be able to see their level... Some people's avatar alignment level reaches a very high tier. And then you feel like this thing is working, it has PMF." 01:42:03

"The users with the best retention are the ones where the avatar alignment level is the highest. They keep continuously adding their context, and then they get more positive feedback — it's a very positive cycle." 01:42:33

Any AI company building a personal agent, companion, or memory product should consider building and surfacing this type of metric to users — it gamifies context contribution and creates a virtuous retention loop.