AI Applications: 100 Questions | How Technology Actually Brings Well-being to Individuals

Image | XinGuang App Illustration

As an early-stage investment firm focused on AI applications, we launched Linear Bolt in early 2024 — a dedicated program for AI application investments (we've already backed 7 startups to date). We pay close attention to innovation, progress, and founder thinking in the AI application space. That's why we're launching a column on Linear Capital's WeChat public account called "100 Questions on AI Applications." We'll invite AI application founders — either from our portfolio or from our network — to share their startup stories and industry reflections, discussing current topics and developments in AI applications. We hope this provides useful reference for anyone interested in the space.

For our first installment, we invited Yuxiao Wang, founder and CEO of XinGuang (心光), a Linear Capital portfolio company, to talk about why he started building this product two years ago, whether he feels the AI + journaling/companion space is getting crowded, and the ever-present PMF questions that AI application founders grapple with.

Image | Apple App Store Featured Recommendation

1. Linear Capital: Please briefly introduce XinGuang.

Yuxiao Wang: We call XinGuang a "life companion" — it's our fresh take on life journaling. Think of it as a life buddy. You can put your thoughts, reflections, interesting moments, meals you've enjoyed, movie reviews — everything — into XinGuang, forming your own book of life. Inside XinGuang, your AI friend shares life with you. Our users configure their AI friend as a bestie to swap stories with, a favorite pop star, a wise life mentor, and other roles, bringing warmth and companionship to their daily lives.

Image | User Feedback

2. Linear Capital: Why did you build XinGuang?

Yuxiao Wang: I started as an indie developer and built a product that reached over a million users worldwide. That experience taught me what it means to craft something small and beautiful. More fundamentally, the core of building a product is answering one question: "What pain point do you actually have, what problem do you want to solve?" XinGuang came from a major pain point of my own. Before this, I enjoyed writing — I'd accumulated over 400,000 words — and my biggest frustration was the constant stream of fragmented ideas scattered across different note-taking apps, impossible to find when I actually needed them. Typical note apps let you search, but they don't help you use what you've written. My co-founder had been journaling for years, on Douban, on Weibo, and her frustration was: she'd written so much, these apps should know her well by now, yet traditional apps never organize or distill your information. So we wanted to find an app that could automatically sort, analyze, and summarize. We couldn't find one, so we decided to build it ourselves.

When you want to build something like this, you need to figure out if others need it too. I did user research, talking to nearly 100 people. I expected people to be averse to journaling, but over half of those surveyed actually used text to organize and express themselves — tools ranged from Xiaohongshu, WeChat favorites, Notes, even pen and paper.

Synthesizing the research, I bucketed user needs into several categories:

First, recording matters deeply to many people — memory fades so easily.

Second, some keep "success journals" — looking back at past achievements for comfort during hard times.

Third, emotional release — when you're sad, you need an outlet.

Fourth, wanting your authentic inner self to be seen — needing a genuine channel for self-expression. These days, people talk about posting "renovated" moments on social media while living "raw" lives offline. With social pressure and the gaze of others, authentic self-expression becomes genuinely difficult.

About two years ago, our idea was to build an app that truly understands users and helps them organize and analyze — this had two components. First, data science: structuring and categorizing data, with a recommendation system to surface content at the right time. Second, AI to build indexes and organize everything. At the time, GPT-3 was already showing glimpses of speaking "human" — we saw this as a direction that would develop powerfully. And so XinGuang began.

Looking back, our initial vision was fairly simple — we wanted to achieve three things:

First, automatic categorization and organization of everything you write.

Second, making it interact like a friend, exchanging life with you.

Third, true personalization — since everyone's life differs, it should serve different people differently.

Image | XinGuang Product Feature Introduction

3. Linear Capital: You started two years ago — were there many AI application developers back then?

Yuxiao Wang: We began this project in May 2022. There weren't many people around me building AI applications then. ChatGPT was the real milestone that brought AI to everyone's attention.

4. Linear Capital: Did large language models provide a massive boost for building XinGuang?

Yuxiao Wang: Yes. We'd been waiting. I mean, when we started this project we knew this would be an interesting direction, and a question we constantly asked ourselves was: "Is your product capability coming from GPT, or from the product itself?"

Looking at AI's trajectory, its real development won't be measured in just one or two years. At the emergence of large language models, I had several predictions:

First, GPT represents the aggregation of all knowledge. Books contain knowledge, but if you don't read them, they're meaningless. GPT has essentially read all books, possessing all knowledge. Knowledge requires logic — which brings my second prediction: strong logic. Humans need their own logic to string knowledge together before it becomes useful. Third, with the logic to connect knowledge comes intelligence. Fourth, only with intelligence can there be what we call emotional intelligence — the ability to perceive how you're feeling at a given moment and understand why. When models become emotionally intelligent, we felt GPT-4 marked a meaningful milestone.

Thinking through these four points, we also asked: what will AI become in the next decade? We started pondering this two to three years ago and concluded it would likely be two things: First, whether humans want to admit it or not, AI will be smarter than us to some degree — what Ilya calls superintelligence. Simply put, more intelligent systems will become foundational; poorly performing models will be eliminated because they have no reason to exist. Second, multimodality is another exciting frontier — models will definitely be multimodal. For some, this may be painful, fearing job displacement, but every technological revolution brings this, while also dramatically amplifying human capabilities in many areas.

So when building this product, we asked: with ChatGPT, we anticipated its strong capabilities, so what should we actually build? Two approaches: first, wrapping — doing whatever it does; second, building something the large model can't do, while our capabilities benefit from it. We chose the second, stemming from our vision — ultimately we want to bring something good to users. My co-founder previously worked on algorithmic recommendations at a major tech company. Algorithms are neutral; today's large models are too. Some use them for deepfakes; we want to use them for positive change.

Image | XinGuang Product Feature Introduction

5. Linear Capital: From inception to now, what have been XinGuang's key development milestones, and what was happening in the industry or at your company at each point?

Yuxiao Wang: Technology itself is neutral. We've always hoped XinGuang could bring the world's best technology to users, letting them experience the happiness that technological progress brings.

For example, completing our multi-platform development was a meaningful milestone. This wasn't industry-driven; it came from our understanding of user needs. We invested significant time developing for desktop, watch, mobile, and other platforms, with the goal of making self-expression possible anytime, anywhere. The AI industry moves fast, especially on the model side, and our approach there is to embrace change — voice capabilities, image recognition, and so on. Each improvement on the model side has been an important milestone for us, and we've kept the product current.

6. Linear Capital: What features do users particularly love?

Yuxiao Wang: One beloved feature is "Letters" (来信). Users configure it in different ways — for example, setting it as Mayday, so when "Ashin" sends them a letter, they're thrilled.

I see two product categories right now. First, tools like Notion with strong recording functionality — Notion's AI is smart, but it's all about you, with no entertainment value. Second, products like Character.AI for chatting, almost gaming-like roleplay. But many users find a pain point: during roleplay, they can't express themselves, or with Character.AI, you're constantly switching between playing different roles.

So we wanted to combine these: your own records — fed to GPT — ask it questions and interact, get key information you want, and automatically save it back. This pipeline happens automatically in XinGuang. After self-expression, you can have deeper conversations, and the results auto-save.

7. Linear Capital: Have you noticed more apps in the journaling space? Some argue that human memory is data that doesn't exist on the internet and that large language models can't access. Additionally, the physical world can also collect data not found online. Why do you think everyone is pursuing these two directions?

Yuxiao Wang: There are definitely many products in this direction now. When we started, few were looking at it. This comes back to: what's truly scarce in this era? When we have genuine superintelligence, what remains truly scarce is what's inside each person's mind — and how valuable is that? I don't think it's necessarily measured by quantity. What's in each person's mind is most valuable to that person, which in turn reveals the value of the individual.

On this point, if we can leverage each person's own data to provide more value back to that individual, I believe that's meaningful.

Imagine ten years from now, when superintelligence may surpass humans in many ways, embodied intelligence included — where does human advantage lie then? We bring nothing at birth and take nothing at death; what truly matters is the experience of life itself. Within that experience, how can technology bring happiness to the individual? That's the fundamental question of our industry. We named the product "XinGuang" — "light of the heart" — because we believe in individual uniqueness, and we want the product to explore using your own data to nourish yourself, making yourself better and happier.

Image | XinGuang Product Feature Introduction

8. Linear Capital: Note-taking and companion products seem to have more granular directions — some lean toward psychological counseling, others toward play. What categories do you see, and what's XinGuang's specific positioning?

Yuxiao Wang: I don't think there's a standard answer for AI companionship yet. Everyone's experimenting in different directions. Early products like Pi aimed to train an emotional model. Some approach it from the angle of human memory banks. Some do roleplay. Some focus on emotional release.

XinGuang is doing life-dimensional companionship. More specifically, we're seeing users who've been with the product for one or two years and continue using it. People use it hoping to find friends, talk to idols, seek life guidance, and so on. This companionship operates on a longer timescale. Games provide short-term stimulation; XinGuang is slower to warm up, but once you're in, you discover genuine tool value. The more it knows you, the more it can do. When Pi launched, we were especially curious whether it would achieve major success. There are many such products, and many have died, but they're like grass — one crop grows, another withers. Who's chosen the right direction? I still think it's unclear.

9. Linear Capital: So can we understand that your product's specific purpose is relatively open-ended, allowing users to engage with it motivated by broader meaning rather than needing a very specific use case to open XinGuang?

Yuxiao Wang: Yes. But my understanding is: when users come, they may have emotions or not — both are normal. They come simply wanting to record something, an idea, a fleeting inspiration. In 2023, we built an internal judgment system to determine whether the user has an idea or an emotion at that moment — these are completely different things. If there's emotion, interaction may be needed; if they just want to record, too much emotional response feels annoying. So it's essentially designing a system to determine when this AI friend should engage.

10. Linear Capital: Setting boundaries matters enormously for such applications — perhaps what you choose not to do matters more than what you do.

Yuxiao Wang: Yes. In XinGuang, chat dialogue was only added in July 2024. Large language models have been around so long — adding dialogue wasn't a capability question, but a choice. And even now, it's not a true conversational chat system; it's designed to help users extend previously started topics.

Image | XinGuang Product Feature Introduction

11. Linear Capital: What are your short-term and long-term plans for XinGuang?

Yuxiao Wang: XinGuang's direction is fairly clear. Our roadmap: in 2022, we were building the MVP (Minimum Viable Product), getting it to users quickly and iterating. In 2023, we focused on multi-platform — now all iOS platforms are supported, Android is in development and coming soon. Why multi-platform? Because for life products, we want users to be able to quickly drop in thoughts or records whenever, wherever — convenience matters. This year, we've increased the companionship emphasis, building many new AI companion features. We feel the infrastructure is ready and want to go deeper — for example, in the coming months we'll continue building out the companionship layer. Platform-wise, XinGuang is quite complete: overseas iOS and web matter significantly; domestically iOS and Android matter, with Android possibly more important.

12. Linear Capital: How do you view PMF, and what metrics do you prioritize? Any data you can share?

Yuxiao Wang: On PMF — I think for products in this general form, including XinGuang, nobody has really found it yet. Character.AI genuinely has. Everyone's still in exploration mode. PMF is a process, but you'll definitely know when you hit it. We're currently at over 100,000 users — not a huge product. Much of our functionality requires payment, and our business model needs adjustment. The current model was designed two years ago; we're considering shifting to subscription, because in the AI era, one-time purchase feels inappropriate — like trying to prepay a lifetime of electricity when you move somewhere. We're also heavily focused on going global.

Image | XinGuang Product Feature Introduction

13. Linear Capital: How did Character.AI's recent acquisition affect you? What do you think it means for AI startup development and exits?

Yuxiao Wang: I think Character.AI's acquisition was a good outcome for them. Their model is genuinely expensive — serving many people at scale requires guaranteeing compute to maintain model quality. I deeply respect them: they had original ideas, were among the first to forge this path and make it work. Many products exist, but few are truly original. Character.AI is among the rare products that defined a category.

For XinGuang, we still hope to build a product that serves more users, brings them more value, and improves their lives. I want to spend ten years on this, making it a genuinely influential product.

14. Linear Capital: A closely watched topic in AI applications is when the killer app will emerge. What do you think? What do AI application founders discuss most?

Yuxiao Wang: I think several killer apps have already emerged: ChatGPT, Character.AI, AI search products, and domestically, Moonshot AI. But if we map this to internet and mobile internet eras — 2G MMS for photos and voice, then QQ as a national-level product, then WeChat, Uber, Airbnb, e-commerce products, and by 5G, Douyin and others. Every era sees several stunning products emerge each year. By that frequency, AI is accelerating. And I think the AI era will be even more fragmented — countless apps will emerge, because every industry deserves to be rebuilt from scratch. Many products and new things are currently at the starting point of being rebuilt; talking about returns now is far too premature and impatient. Something new always needs time to reveal what it actually is.

Every era has its own kind of fortune. The current era is a great time for recent graduates who want to build something, especially AI-related, because this era has just begun, and we can be certain AI will be among the important things of our time. People worry today whether AI is overheated — I think it will overheat, because good and bad actors alike will come explore and experiment in different directions, like a gold rush. Gold has always been scarce, but with more people digging, the probability of finding some increases.

15. Linear Capital: Some AI developers say investors look but don't invest, and some full-stack developers could build a "one-person company" on their own. Why did you choose to raise funding?

Yuxiao Wang: I previously taught iOS development to entrepreneurs and interviewed many people, and one thing struck me deeply: it's not that one person can't build a company, but true generalists are genuinely rare. Take language — study Chinese long enough, and your brain gets "optimized" to that language's expression patterns. People also easily label themselves: "I studied X, so I can only do Y" — this mindset is common. AI, conversely, is a generalist that can compensate for human limitations.

On fundraising, our thinking is quite simple. In this era, some products can thrive without funding — that's fine, but seeking investment in that situation makes both sides uncomfortable. I've built indie products before; I know how to make money and generate revenue. For us, fundraising mainly comes from our starting point: we want to make a ten-year attempt, seizing current opportunities to see if we can do this well. Funding gives us more confidence. Survival is magical — like the gold rush analogy: everyone is panning for gold, and whoever brought more water gets to dig longer.

---

Linear Bolt

Bolt is Linear Capital's dedicated investment program for early-stage, globally-oriented AI applications. It upholds Linear's investment philosophy, focusing on technology-driven transformative projects, helping founders find the shortest path to their goals. Whether in speed of action or investment approach, Bolt's commitment is lighter, faster, and more flexible. In the first half of 2024, Bolt invested in seven AI application projects including Final Round, XinGuang, Cathoven, Xbuddy, and Midreal.