#246 - Agnes AI: Southeast Asia's Answer to ChatGPT (And 20x Cheaper) - Bruce Yang

[00:02:02] Introduction

Henry Suryawirawan: Hello everyone. Welcome back to another new episode of the Tech Lead Journal podcast. Today I’m very excited to have someone who is closer to my presence, right? More local than the usual guest. His name is Bruce Yang. Bruce is actually the CEO and founder of Agnes AI. So before this, I haven’t heard a lot about Agnes AI, but it’s actually one of the, you know, top most growing, AI model and AI app at the same time, coming from Singapore. So it’s much more local here, than, you know, the ChatGPT, the Gemini, the Claude, all the models that we hear a lot, recently. So I’m really excited to talk a lot more about this local model. And yeah, Bruce, welcome to the show.

Bruce Yang: Thank you very much to have me, Henry. Very excited to share about Agnes.

[00:02:49] Why Did Bruce Start a PhD During COVID to Build an AI Company?

Henry Suryawirawan: So Bruce, in the beginning, I always love to ask my guests, maybe share a little bit more about you, by sharing any career turning points that you think we all can learn from that.

Bruce Yang: Sounds good. So I’ve been studying in Raffles Institution for my high school. And after that, I’ve been studying in the US at UC Berkeley for undergrad. And right after that, I’ve been working in the Valley for over four years in Microsoft and LinkedIn. And after that, I’ve been doing startups in the Valley and decided to come back to Singapore, actually right before the COVID. So during the COVID, I, there’s actually nothing much I could do. So one thing which I think is a turning point for my career is I started my PhD during COVID at National University of Singapore. And I think this is a very good timing for me because, you know, it’s good time to think about the second curve of my career. And it’s right before the start of the ChatGPT.

So during my study and my research, as a PhD, you know, a researcher at NUS, we’ve been very well exposed to all the brand new stuff like the agents, like large language model and like the memory, agentic memory. And there lays the foundation for the starting point of Agnes AI and which I think is one of the most exciting moment of my life. So I’ve been talking to a lot of people, seeing the opportunity of Agnes and any exciting consumer app of AI for recent years, if you miss this for once, you miss it for your life. So definitely getting very much excited about, you know, founding this company. And I think that my PhD is a very important turning point for me. Yeah.

Henry Suryawirawan: Yeah. So I think you brought up a very interesting point, right? If you miss this, you know, moment in time, right? You kinda like missed the, you know, so-called the turning points in the world, in fact, right? And I mean, as a user, even like this is also like a defining moment for a lot of people getting used to working with AI, knowing AI capabilities.

Bruce Yang: Right.

Henry Suryawirawan: I can’t say for sure, like, by being a founder, you know, someone who is building an AI model and AI app at the same time. I’m sure this is must, this must be more exciting for you.

[00:05:13] Sweep Sponsorship

Henry Suryawirawan: Before we continue, I want to tell you about our sponsor, Sweep.AI

I love my JetBrains IDE. And if you’re like me, you know the struggle. When it comes to AI coding assistance, we’ve had to compromise. Maybe alt-tabbing to other tools, or settling for autocomplete that just doesn’t cut it.

Well, that changes with Sweep AI. Sweep is built specifically for JetBrains IDEs, and honestly, it’s been a game changer for me.

The auto complete feels instantaneous. The AI actually understands your code. And there’s an agent that genuinely helps you write code faster. Plus, your code stays secure.

I finally found an AI that works great in my JetBrains IDE. No more jumping between multiple tools just to get decent AI assistance.

You can download Sweep for free at sweep.dev. Thousands of developers at companies like Ramp, Amplitude, Atlassian, and Klook are using Sweep every day. See for yourself why they’re loving it.

And now let’s get back to our episode.

[00:06:16] Why Build Another AI Model When Thousands Already Exist?

Henry Suryawirawan: So maybe the first place I would like to ask, because starting from ChatGPT, all this craze about AI, LLM, generative AI and all that, right? The last I checked, you know, in Hugging Face or maybe some internet articles out there, there are plenty of models already available in thousands, hundreds of thousands. Why creating another model with Agnes AI?

Bruce Yang: Well, there are a lot of models, and this is actually the reason we wanna start our own one. The models have a division of two kinds. One is like the SOTA, the closed source models provided by ChatGPT, Anthropic, and maybe Gemini. And another set is defined by the open source ones, you know. One was the very famous ones are DeepSeek and Qwen and Llama, and a few more. We’ve been through a time that, you know, the SOTA models are mostly from the closed source one and open source one are actually lagging very far behind. But right now, the gap is very small. So we think that this is the best time for us to find our way to close the gap between the open source ones and closed source ones.

We have the, all the requirements which we need, all the ingredients for us to do this job because we have a very strong research team. My advisor joined us as part of the founding team of Agnes. I have another friend, a Berkeley alumni who went to MIT for his PhD. He’s part of my research team. And we have a lot of users using our product out there, especially in the Southeast Asia. We have over close to 4 million registered users with something like close to 300K DAU by end of this month. So we have a lot of people using our product every day and that generates a lot of data, a lot of prompts. Using these prompts as, you know, the seeds to train our, post-train our model through a reinforcement learning. We could potentially also distill it, distill the our model from deploying the SOTA models in our production. So there are multiple ways to close this gap, especially we have our, you know, product out there, users using our product, generating all this kind of data, and this is our opportunity. Yeah.

Henry Suryawirawan: Wow! It must be very exciting for you seeing, you know, like 4 million registered users and 300,000 daily active users. That’s pretty a lot, right? So maybe one thing to verify, so Agnes AI is a open source model, is that correct?

Bruce Yang: For now, it’s not open sourced, but we have, you know, published the paper, which details about how we train our model. So it’s somewhat open sourced. The reason we haven’t open source it is we’re preparing for another open source model, which will be released very soon. Some of our smaller models have already reached SOTA, which is state of art for the range of parameters like 7B, 30B. But if you want to really make a noise and be the king of maybe Southeast Asia, we have to train an even bigger model. Right now, we have seen very good results. We have, for the 30B parameter model, we are already reaching very good results for SEA or quite a few benchmarks from the Southeast Asia region. But we want to enhance it slightly more before we release it to the public.

[00:09:48] How Is Agnes AI Cheaper and Faster Than ChatGPT?

Henry Suryawirawan: Alright, thanks for clarifying that. So maybe let’s go into the differentiator, right? So what makes Agnes AI a compelling options for people who are used to, you know, ChatGPT, you know, Anthropic Claude, Gemini, whatever models that are available out there.

Bruce Yang: So if you’re talking about the model itself, we are very much specialized for the work tasks like doing a research, writing a PowerPoint, or maybe something like, you know, designing for a picture or generating a video. You know, we have both the large language model to generate the props and the image and video generating models to, which is the DiT diffusion. Transformation, diffusion transformer model to generate the content. And we also have a model to support our group chat. So you can think that all the models which we trained are specialized for your daily tasks instead of your daily conversations. So this is one thing which we are differentiated from ChatGPT or Gemini or Anthropic.

Another thing which differentiates us a lot is we are cheaper and way faster than most of the big models because we are specialized for more narrow tasks. We don’t need to use very big MoE models. Smaller models will suffice with some proper routing. By routing, we mean that if you know the, what the intention of the user, we would not need to direct the request to the big model all the time. So for example, if you need to work on your PowerPoint, the model specialized for the PowerPoint by writing the HTML content for the PowerPoint, especially trained, which on-par with that, with, on-par with the SOTA models like the Gemini or ChatGPT, but it’s much faster and much cheaper. So this is how we differentiate ourselves from the other models.

Henry Suryawirawan: So that’s very interesting. You mentioned it’s cheaper, it’s faster, right? So obviously these days we don’t actually perceive about latency much. But cheaper is definitely in everyone’s mind, right? So by having a model that is…

Bruce Yang: Right, especially you’re talking about the emerging countries. So one of the observation which we had is, you know, only like 0.5% of users are paying for generative AI products. So this is the data from Manual Ventures. There’s about 500 million to 700 million users using GenAI product every day. Comparing to the netizens in the world which is about 6 billion, is about 10%. And out of this 10%, if you’re looking at the products like the Gemini or ChatGPT, there’s only about 5% of people are spending money for subscription. And this is mostly centered in the high paying markets like the US, the Europe, Korea, and Japan.

And for emerging countries like the Southeast Asia, like Latin America, very few people spend any money on these products. And if you can really reduce the cost for them, reduce the cost for ourselves, we are, you know, kind of unblock ourselves from serving more people. Our goal is to create a product which will be everyday AI for everyone, serving the emerging countries, be very inclusive. We want to keep our costs really low. Much cheaper than that of ChatGPT, Perplexity or even that of DeepSeek or Qwen. So the way we solve the problem is reduce the size of the model while achieving the same kind of standard of the results. So we’ve been doing extremely well for this. We, Our cost for each of the tasks is like one 10th or one 20th of that of all the major products out there. And that give us a huge edge to serve the unserved population in the world. Yeah.

[00:14:00] Does Agnes AI Support Southeast Asian Languages and Cultures?

Henry Suryawirawan: Wow! Very exciting! Like when you mentioned one 10th or one 20th, I think it’s, you know, starts very compelling, right? So one question as well because you kind of like the answer to DeepSeek from Southeast Asia, right? Do you also support Southeast Asia local context, meaning like languages, culture, those kind of stuff?

Bruce Yang: Yeah, of course. So this is a very important thesis for us because if you look at the population in the Southeast Asia, there are a lot of people speaking minority languages. Like Bahasa, like Thai, Malay, these language are not very well served by most of the major large language models because of the corpus, the lack of corpus of data from the literature. So what we can do from here is instead of trying to make a product which is doing every, doing well everywhere, we want to make a product which especially aligned for the local interest, especially serving for the regional languages. And the way we do this is post-train from open source models, but with more data, more language, more literature, and more, you know, data from own, our own platforms, from all the competitions from Agnes. That way we can bring more interesting content like the lens, the spoken languages, to our platforms for people to understand about the culture, understand about the ethnics. That’s something which I think is a huge advantage for us.

[00:15:34] How Does Agnes AI Handle Local Languages Better Than Global Models?

Henry Suryawirawan: Yeah, you mentioned about the lack of corpus, right? So the last I used, I’m Indonesian by the way, so I speak Bahasa sometimes. I also converse with AI using Bahasa and it seems to be able to understand. So you mentioned the lack of corpus here. Maybe give us some use cases or examples where, you know, maybe things like Gemini or ChatGPT cannot serve but with your model actually it can find better corpus, better data. Because I assume a lot of these models just train using, you know, internet data, right? They just, crawl the internet and get all this data, right? How do you train differently, basically?

Bruce Yang: That’s a very good question. So I won’t say that GPT or Gemini are not doing well at all in terms of the minority languages, but because they are using the same model to serve every population, every person in the world, among all the, you know, geographical regions, you can’t really do well everywhere. Especially if you think that the major languages are English, Chinese, European languages like French, German, Japanese, Korean, it’s very difficult for you to serve the major language well at the same time cater to the minority languages with the same model. So that’s how we are differentiated from them. We, it looks like we’re all coming from the same pre-trained model, but we’re giving a more weight, a bigger weight for the minority language. It doesn’t have to be different literature. It could be the same.

And the one ingredient which we start differently is we inject a lot of data from our own platforms. And there are a lot of people using our product in the region. 60% of all our population are from the Southeast Asia, speaking Bahasa, Malay, Thai, and all the languages, and Filipino. So this provides something very meaningful to us because if you only get data from online, from literature is mostly formal language and you don’t use that a lot in your local, you know, everyday communication. And we have a lot of data from this daily use cases. We inject that in our post-training and that just help us to make a better, you know, stand out and make a better satisfaction for local audience with our own models.

[00:17:57] How Does Agnes AI Reduce Hallucinations?

Henry Suryawirawan: Yeah. So one of the definitely challenges when working with this LLM, you know, generative AI is basically about hallucination or, you know, giving the wrong, inaccurate information. So how does Agnes differ? You know, do you face the same challenges or do you do something different to minimize that?

Bruce Yang: Well, the entire problem of hallucination has been dealt much better, especially with the new models. We do a lot of post-training from the pre-trained models like Qwen and DeepSeek. For the newer versions, they’re already performing much better. So we take some free ride. Besides that, we also use a lot of, you know, citations. We’re sourcing our, putting a support from all the sources for all the claims which we provide in our research and search results. And that give us better, you know, fact check for us.

Besides that, we, another way to solve the problem is using multi-agent systems. So sometimes one large language model can be wrong and the model itself as one agent do not really know where the problem is. But it’s very easy to be checked by another agent. So if you have multiple agents, especially multiple agents from different models, looking at the result at the same time, playing different roles. One is working on the generation. Another one is working on the evaluation and verification. The third one is working on providing the feedback for some correction. All three of them working together solve a lot of hallucination problems. So we’ve been dealing with this kind of situation a lot. And we enabled this trio situation of generation, evaluation, and correction. So I think that’s a very good way to solve problems like hallucination.

Henry Suryawirawan: Wow, very fascinating using multi-agent approach to actually deal with this. So definitely it’s something that we all can see in terms of the results, right? Whenever we use any AI model.

[00:20:03] What Can Agnes AI Do That ChatGPT Cannot?

Henry Suryawirawan: So maybe let’s dive deep a little bit more about Agnes AI. What kind of features are available? You mentioned about, you know, daily tasks, you know, maybe like generating research or PowerPoints. Or do you also do it for, I dunno, emails, any productivity things? Like maybe tell us a little bit more, what are some of the core features or the like, competitive advantage that Agnes provides?

Bruce Yang: Yeah, so Agnes is both the name of the model and the name of the product. In terms of product, we think we are, we’re doing pretty well. The number speaks for itself. And we have both two different platforms, one is on the PC and one is on the mobile. The positioning of the two platforms are slightly different.

If you are working on the PC, a lot to do with the productivity. So you can do a deep research on Agnes. With the same context, you can start a PowerPoint or you can generate images or videos and you can inject the video image into the PowerPoint. It can also help you to work on the Excel. It reads the Excel sheet very well. A thousand, you know, lines of record can be read within half a minute. And if we understand the content and the relationship and you can draw the graphs for you within half a minute too. So it solves a lot of problems of heavy lifting. We would call it dirty work. If you don’t have AI supporting you, you might spend hours. I would spend hours, five hours, 10 hours, working on PowerPoint. And you have to start with the research. It’s like a pain in the ass. With AI, it solves the problem very well. And the good thing about Agnes is it has a agentic memory. That means the first day after you use it, it knows where you stop from yesterday, and it will start the task with the context which you left off. And you can also pass a context to your colleague, through a group chat or from a shared collaboration. That means you don’t have to just drop an artifact, like a PowerPoint to him and he doesn’t know what, how you’ve come up with this result. AI would help you kinda understand where all these results are coming from and answer some question for your colleague, for them, for him or for them to pick up from you. So this is on the PC side.

On the mobile side, we are starting more social, entertainment, it’s like more, you know, less formal. We really want our mobile experience to be more inclusive compared to that of ChatGPT. And not only limited to productivity. If you look at the retention numbers, usage hours for different apps, you know, the social apps like TikTok are doing way better than that of the productivity apps like the ChatGPT. GPT, their, you know, daily usage time is like by average like 40 minutes. If you start with TikTok, WhatsApp, normally, hours will be spent.

So I definitely think that when we talk about AI right now, people still limit themselves into the productivity category. And if you look at the entire internet, entire, you know, usage of the internet, right from you wake up until before you get to sleep, you’ll spend most of your time working on communication. Like how we are doing the interview here right now, on the WhatsApp. And also you spend a lot of time at your free time to look at, you know, entertainment, or contents and things like from Instagram, TikTok, maybe Facebook.

So we think that all this content can be very well supported, assisted by AI. It’s just that nobody have done that just yet. And it’s very difficult for GPT or for Gemini, for any other products to do the same thing because they have already have a very strong user base. And if they wanna make any change, they have to start a new app. So we are able to integrate all these social features into the productivity category because we’re still early to the market and we want to really cater to the young audience, the ones which we call the AI native population, the ones which, you know, had their high schools or college while during the COVID, stuck in their own room and, you know, have to jump on the virtual world. Right after they get, you know, unblocked, the, it’s the year of, the burst of ChatGPT. So they’re very native to AI. Any product without AI seems odd, seems banal for them. So I think this is something which our product will help, will get a chance. If we are able to serve this AI native generation, if we get them used to using our product instead of that of WhatsApp or Snapchat, we get a chance to become the next Meta. That’s the objective behind it.

Henry Suryawirawan: Very exciting! I still cannot, you know, wrap my mind around, you know, productivity and entertainment at the same time, because I feel those two are opposing things. It’s like one thing distracting from the others. But definitely will, there’ll be a lot of innovations looking forward for Agnes to do that.

[00:25:31] Why Is AI in Group Chats the Next Big Thing?

Henry Suryawirawan: So one thing in particular that you seem to focus a lot is the group chat, right? I rarely use an AI that is within a group chat. So one thing that I know is like in WhatsApp you have this Meta AI. Maybe in Telegram you have bots, right? But I rarely see an AI that actually participates in a group chat together with you and few other people, right? So tell us why this is the next thing that people should know about and use more. What kind of things that these AI can do in a group chat?

Bruce Yang: Well, so, you know, we are not the first players to include group chat in AI. ChatGPT has just launched their group chat recently. And Snapchat as an incumbent player, they are working with Perplexity. I think they’re gonna spend something like 500 million to gather services from Perplexity to support search in their chats. So I definitely see the trend of, you know, AI and chats working together. And the chat, which we’re meaning a real chat, not a chat with AI. I mean, like the ChatGPT.

So, you know, AI in the chat can be very disruptive - I mean very innovative because during your normal chat with your friends, there’s a lot of loss in memory and loss in context. If you miss a communication in the morning, you would spend a lot of time to look at all the communication in the morning but it may not really contain a lot of information. And AI can help you summarize the content and that it jumpstart with a topic. First of all, AI serves as a common memory for all the people in the group. Second, you know, sometimes people won’t really communicate very well. There’s a lot of misunderstanding. AI sometimes can understand people better. One sentence may mean two different things to different people, and some of people are very introverted and other peoples would be more expressive. We would hope that AI would fill the gap to make all people very expressive. If the person can really express for himself, AI will fill the gap.

There are a lot of fun things. One of the things which we are thinking about is like a game. Some of the AI games in the group chat, like, the Netflix show of Squid Game, right? Squid Game. We’re thinking about, you have to convince AI to not kill you, but kill another player during the session. And every once in a while, once for while, you know, half of the people will be muted until you find a winner. And there might be a some reward like in our product we are giving the credits. So we give like everybody, we’re committed for 10 credits. And at end of the game only the two players or one player will get all the credits. So these kind of cool things, fun things would only be enabled with AI. And we’re also thinking about other games like guessing who is human in the group chat. The AI have to act like human or human have to act like AI to misguide other players. So this kind of cool things is very well enabled because AI has gone into a stage that is very similar to the logic of human when they speak, almost passing the Turing test. So a lot of cool things, fun things can happen in a social scenario, and that’s something we want to, really want to explore.

Henry Suryawirawan: Definitely, those are fun games to try. So if people are interested, I guess they can sign up to Agnes AI.

[00:29:18] How Does Agnes AI Keep Your Private Group Conversations Secure?

Henry Suryawirawan: So you mentioned about there are a lot of context missing in a group chat or misunderstanding happening. You know, it’s trying to also be more inclusive. And especially with the Southeast Asian context, right? We are kind of like different for different countries, right? Different languages, different culture, different- even like the way we speak is different, right? So I guess it will help a lot if we have this kind of AI to help us to, I don’t know, smoothen the conversation, make sure no misunderstanding happening. But what about the privacy that is ongoing between all those chats, right? How does Agnes actually protect privacy and security for these kind of conversations, yeah?

Bruce Yang: Well, so one thing which we can support compared to all, compared to all the other, you know, application apps is we have our own models and we have our own, you know, database. So that means we are able to solve the problem from our own end instead of playing, you know, switching between different apps, different models, and that could introduce a lot of complications. So privacy is a huge problem for us. I mean, a very important topic for us. We are working with all the, you know, GDPR, all the guiding rules for different regions.

If you, we talk about, you know, enterprise solutions, we’re able to, you know, deploy our services, our clients, you know, their own servers. They also reduce any risk of information leaking. So I would say, you know, besides all the important measures, we can handle this slightly better because we have our own models and we have our own database. And that means we don’t have to deal with, you know, data exchange problems. But anyway, I think AI would definitely help with a lot of governance in the community, in the social media, in order to reduce any kind of data leaking. Yeah.

Henry Suryawirawan: Yeah. So you mentioned that someone can host Agnes model inside their premise. Is that correct?

Bruce Yang: That’s right. That’s right.

Henry Suryawirawan: Okay. That, yeah, definitely that’s one also cool thing, right, to protect privacy and security. Especially if you deal with, you know, I dunno, like very sensitive data or highly confidential data, yeah.

[00:31:41] Will AI Make Us Lose Our Critical Thinking Skills?

Henry Suryawirawan: You mentioned that people can get help from Agnes AI to produce PowerPoints and documents, research, images, and all that, right? So definitely that’s a like a huge help for productivity and all that. But the other aspect of that is like some people are concerned about the critical thinking aspect that might be affected because we leverage AI or we just outsource everything to AI. You know, AI will produce everything for you. So what is your thought about this critical thinking aspect and anything that you think Agnes AI do differently to help users not to, you know, like soften their critical thinking, but actually also collaborate together?

Bruce Yang: Yeah, I think Jensen Huang has mentioned that in the, you know, symbiosis of human and AI, human will need to give a lot of inspiration while AI doing the heavy lifting work. I don’t really think that AI would serve the purpose of, you know, introducing all the inspiration and entire end-to-end work for human. Because if you look at Sora, it’s a huge success at the beginning, but nobody is using it right now because everything is AI. It is, there’s no real human content in it. And the density of human, UGC, user generated content, will be very important success for community because it’s like a mutual respect. If you, this is content generated by AI, why do I need to read it? If it’s content generated by human, there’s a lot of compassion in reading the content.

So I would say that Agnes serving as a product of AI native social app, you know, integrated with the productivity. We will value a lot into human inspiration, human content, and we put human in the loop at all the major milestones. So for example, if you really want to produce a very good PowerPoint, you would do a very thorough research yourself, and you would go for multiple rounds of prompting to get the content you needed. And you would really ask Agnes to help you look into the content, verify the content and put up something like, page by page, you know, layout. And we enable people to have multiple rounds of communication, to edit the content, edit each of the page content before start generating of the HTML, which is the end part of the slides.

And even the first round of slides are generated, you are able to edit very well. We support, you know, an advanced editing function, which is almost like manual editing on PowerPoint, but it’s much more smoother. And you can also have Agnes help you change the page. Like change the logos, change the image, change the background. By the end of the process, we would hope that there’s very little, you know, signals of AI. Very little, you know… You won’t really see the AI content within these slides because AI is doing everything which human wants it to do, instruct AI to do. So this is how our philosophy of the product is changing the development of all these features. Yeah.

Henry Suryawirawan: Wow, I think that’s very important, right? We don’t wanna lose our critical thinking at the same time leveraging on AI. But definitely AI helps to boost a lot of productivity, right? And in one aspect, it actually can boost our critical thinking as well. If, let’s say we provide a lot of inspirations, just like what you said, right? So if we provide a lot of inspirations, yeah?

Bruce Yang: Yeah, so this is nothing to do with Agnes, but I want to also wanna share a story about myself. While I was doing my PhD, one of the classes which I took, the robotic class. I was A+ student, A plus like top student in the class. And the reason I was able to do that, well, I started my PhD not right after my undergrad. I’m like 10 years older than all my peer classmates, but I’m still like the number one in the class, which is very surprising. And the secret which I share to all the, all my friends when they ask about the reason is I use a lot of AI, use a lot of, you know, to use like the NotebookLM, ChatGPT to help with, you know, curating the content of my paper and doing the research, doing a survey. And this is all, you know, very well supported by our professor. He, starting from the first day, he’s very open-minded. He said that you should use AI tools, just let us know. If you can generate a very good results with AI tools, you should share that in the class because everybody should know about using AI tools to boost their creativity, boost their productivity.

But one of the things which we don’t really wanna see with this kind of AI, you know, facilitation is if AI is diluting the entire community, that could be a huge problem. Like right now, the Sora app, it used to be a very, very promising product. But if you’re looking to AI, Sora right now it is kind of chasing people away because all the AI content is all looks the same. It’s, there’s homogeneity, which means all the images, all the videos are looking all look the same. Very AI look. So we don’t want this to be the same kind of situation when we involve of human productivity. For example, if you don’t put an inspiration, if you work on a paper, it will all look the same. And if you work on a survey, it will give a very, very general responses instead of insightful, you know, critics. So that’s my comments about this topic.

[00:37:43] Should Children Use AI for Schoolwork?

Henry Suryawirawan: Thank you for sharing your personal journey with AI during your PhD study, right? So which brings me a very interesting topic, right? Because I think there is a dilemma in, you know, leveraging AI for students, right? Especially as early as maybe primary school and secondary school and things like that. What is your view about this? Are you more pros towards leveraging AI even early in the education or do you have a different thoughts?

Bruce Yang: So my daughter, she is exposed to AI because I’m an AI entrepreneur. But I also give her a lot of heavy lifting schoolwork, like the Math Olympiad. She just got in the gifted program in Singapore, which like top 1% of the batch. And she, at age of nine, she has a vocabulary of 10,000, which is quite a lot. She does all the Math Olympiad. She was like top 40 national level. So I definitely see the advantage and the benefit of training the intellectual without the assistance of AI at early stage. Because if you start to rely on AI, a lot of things can be very easy, but you would just lose the kind of habit or lose the critical thinking or you might not even know how to prompt.

That’s the problem. If you know the end results, you know, human analysts have to be the evaluator at the end. You know, what kind of result would be a high quality result. And the only way to appreciate, to have the kind of taste is going through the process manually without dependence on AI. Even though I know the benefit of, I know the disruption of AI in education, I still instruct my kids to work really hard to be a critical thinker. To understand about, you know, rudimentary logic and high quality critical thinking because this will be very important prerequisite for you to be a AI adopter, for you to really highly utilizing AI in your future work. That’s my belief. Yeah.

Henry Suryawirawan: Very interesting takes, coming from a, you know, an AI founder, right? AI entrepreneur. So definitely maybe in early age, you should not leverage AI too much, right? And what you mentioned is definitely very true, right? So if you don’t know how to prompt, if you don’t know what good results, good quality look like, I think that’s definitely a big problem, right? Even though you have AI, but if you cannot produce high quality results, that’s also kind of like defeating the purpose. So I think thanks for sharing your take, yeah.

[00:40:27] Can Agnes AI Help With Coding Like Cursor?

Henry Suryawirawan: So how about coding? Because a lot of use case these days, you know, using AI is for coding, right? So you can generate code front-end code, back-end code, whatever code, right? So is this something that Agnes also can support?

Bruce Yang: We could support coding, but we do not support it in a formal way. That means if you ask Agnes to code for you, maybe give you some code snippets in the response, you would do it well, just like how the ChatGPT is doing. But we do not provide a function for you to code within an IDE just like Cursor. There are reasons for that. I can share in a bit.

Coding is very difficult task. It is one of the tasks which you see a huge difference between the levels of smartness of AI. That’s why, you know, Anthropic is doing extremely well in coding despite all the other, you know, AI models are catching up on everything else.

So we don’t think that it’s very good category of topics or area that we should focus because that means we need to maintain a top level of the model all the time. And that could be very competitive. If we talk about product point view, we want to, you know, for some, a short period of time, we want to be a SOTA, best in the world. But it’s very difficult to maintain at the top level all the time. And if we build a product which sits on the dependency of a SOTA model, there are very few, you know, choices. And we have to be very cautious about making that of decision.

So right now, you know, a lot of other functions like research, PowerPoint, you know, image generating or video generating or maybe a group chat. Even if our model are not SOTA, you might not see the difference, number one. Number two, you won’t really appreciate if we change to a better model because the network effect, the memory, the habits, the habitual use cases would help you to stick on our platform. This is something which we definitely think that will make a better fit for us at this point. Yeah.

Henry Suryawirawan: Yeah. So, yeah, definitely, very difficult right, to compete in coding. We have so many benchmarks, right? And I think over the time we will see eventual winners, but it seems like these days there are a few options only.

Bruce Yang: Yeah.

[00:43:07] Will Everyone Host Their Own AI Model in the Future?

Henry Suryawirawan: Yeah. So definitely it’s one, one as yeah, very few options.

So you mentioned about enterprise that can host AI models. For consumers, right? There are plenty of options. You can subscribe to any models you like. You can even use an open source model deployed on your laptop and all that. And there are again, like there are hundreds of thousands of models available out there. So what do you think the future looks like, right? Because as an end user, I might be overwhelmed with a lot of options. I might not know what’s best for what use case. And many of these things are over the internet, right? So that means I will send a lot of my data. And especially if you use it for, you know, like private conversations, you know, confidential information, there will be a lot of leakages happening. So what do you see the future of this? Do you think everyone in the end will have their own model host, hosted by themselves?

Bruce Yang: I don’t know about that, but, you know, most of the models won’t really exist after like three to five years because the only reason they are existing right now is there’s not a huge dominance and we are sitting on very shaky land, that everything is disrupting. Just like how the ChatGPT is focusing a lot on all different kind of things. And we just see that the model itself is not a huge barrier for a product to grow down the road because, you know, one year ago ChatGPT is still like number one in the world. Within the last one year there are quite a few new players which is competing with ChatGPT like Anthropic, Gemini, and xAI. Within the, you know, half a year from now, GPT might not even be the best model in the world. Right now after the release of Gemini 3, everybody’s saying that Gemini is doing better.

So a good product definitely has to be a, you know, a merge of both a good model and a good traffic, huge used product. And that means that it’s not about the model which make people, you know, doing the choice, I’m making the choice, it’s the product which providing all the features, utilizing all the models, which get people stick on it. So that’s why I think, you know, ChatGPT is definitely doing well. There are quite a few other players like maybe Agnes, maybe us, maybe a few more. But most people won’t be really knowing what’s the model behind it. Because the product, the features, the use cases would definitely be more prominent for people to understand, to resonate.

Yeah, so in terms of whether people would have a product built by their own model down the road in the future, it’s a potential. I don’t know. I think it is pretty wild imagination. It’s something very different from what people are using the apps right now. But it has brought a lot of new things. I won’t say it definitely won’t happen. But as mentioned, I don’t think that there are going to be a lot of models down the road. There are very few models supporting very few products which will dominate in the world. And all the other smaller models will be on the long tail. Very few people use it. It might still exist, but it won’t make a huge impact.

Henry Suryawirawan: Yeah. So I think everyone now is competing, you know, like all the big boys, right? Trying to be the best model available out there. And I personally see also like these AI models seemingly doesn’t have a lot of moat, right? So like what you mentioned, right? It’s not the model itself that makes the whole difference, right? It’s the application of it, the integration with the app, product, the use cases. And I can, like for me myself, even though I sometimes see those extremists trying to like compare one model versus the others, I personally find it less compelling for me because, yeah, I just use auto in some of the products, you know, they’ll choose the best model for me and I’ll just leverage on the output instead of, you know, tweaking which model I should choose based on, you know, what temperature, what parameters and all that. So I think what you mentioned is definitely very, very true.

[00:47:39] Is AI a Bubble or Real Economic Transformation?

Henry Suryawirawan: So I am a little bit interested in when you say that a lot of models will die, right? So I know some people these days are talking about AI bubble.

Bruce Yang: Right.

Henry Suryawirawan: So there are a lot of investments, you know, high valuations. Do you think this is an AI bubble or this is something that can still continue for foreseeable time?

Bruce Yang: So it’s definitely not a tulip bubble, which there’s no substance at all. AI is something which is very well supported by real economic impact. And if you look at the growth of ChatGPT, it becomes so widespread, not because it’s a bubble, but because it does really make some, you know, use cases. I’m using ChatGPT to save a lot of time when I trying to write a paper, trying to write a, you know, email. It give me a lot of very good support. Same from our product, Agnes AI. So with AI support, a lot of people are empowered with making a stronger impact. And this is something we try. I definitely see that it’s a huge, very important growth for the economy.

That’s why I don’t think it is, it, long-term speaking it’s a big bubble, but within a short period of time if you look at, you know, the transitioning right now, a lot of money is betting on the models, and just as we mentioned that, the application definitely will win the game down the road. So for the companies which raise a lot of money also wasting a lot of money making the models but not really being wide adopted by a lot of users, they might be stand at the center of the bubble. But the wealth, the value of AI will definitely grow all together. Just transitioning from the model companies, just we’ll see the transitioning from the model companies to the application companies. And the application will be very universal, not only limited to the large language model, but also like the robotics. Like the enterprise, the automations. So that’s my belief.

So in short, I don’t think that AI itself is a bubble. But micro level speaking, maybe some of the model companies because raising so much money at the beginning, but because of competition, because of the application traffic problems, they might be a bubble.

Henry Suryawirawan: Interesting take, definitely, right? And I recently read this, Southeast Asia economy report. So, you know, every year, you know, they came up with this report, right? And one aspect that is very interesting in the report is definitely the introduction of AI into some of these applications within the region, right? So definitely it’s one major variable in the economy growth for Southeast Asia at least. And I believe it is happening also in the whole world as well, right, in bigger parts of the world. So definitely AI seems to disrupt every industry, I would say maybe, if not most, industries, right? So I think definitely is something that we can foresee for quite some time. AI producing a lot more values integrated with so many other products.

[00:51:01] How Can Southeast Asians Start Using AI Today?

Henry Suryawirawan: So which brings me to the next question for us individuals, right? I know that these days, many people already talking about AI. But I can still see some people are not into using AI yet. So maybe they are scared about hallucinations, they’re scared about AI governance, security and all that. So, and especially for Southeast Asia, I’m sure there are a lot of other people within this region that are not tech savvy, right? So they don’t know what is AI, they don’t know what it’s capable of. What is your advice for us to be more ready, more integrated with AI in our day-to-day life?

Bruce Yang: Well, this is something which Agnes is trying to do. One of the problems with people not using AI is it’s only limited to productivity and is there’s a lot of, you know, cool features which is only provided by the higher paid tier. And Agnes is trying to solve the problem by provide a lot of cool features at paid tiers of other products for free. So for people of the region who wants to understand about AI and try AI, you should try it with Agnes today. I think we are one of the products which provide all the features from ChatGPT, from Perplexity, from Gemini, up to one product. And we provide a lot of free quota. Almost, you don’t have to spend money to enjoy your use of the deep research, PowerPoint, image generation and video generation a lot.

Beside Agnes, you can definitely also try all the other products like Perplexity and GPT, try to, you know, integrate your life, your work, your studies, your communication within, you know, AI scenario. Sometimes it looks a little bit tech savvy, I mean. But I would definitely say that it’s very easy to enter the door because the communication with AI is not using a, it’s like agree. It’s like command, it’s like natural language. For a lot of other tools like Excel or Adobe Photoshop it’s very difficult to understand all the instructions. But AI, you just have to be a natural language speaker. You just have to know how to prompt. So it is not difficult at all.

Henry Suryawirawan: Yeah. So I think the cool thing about AI, LLM, definitely is the natural language interface, right? Which makes it easy for any people, right? Even like with the local model, like Agnes, maybe in their language. They can just speak to the AI as natural as speaking to a person. And yeah, hopefully they can get good results, right? Especially if it’s like a common problem, right? So I’m sure the answers will be much more accurate and in a good standard. And yeah, hopefully we can see a lot more people within this region, being AI trained, being AI enabled, and hence impacting the economy a lot.

[00:53:56] What Are Real-World Examples of People Using Agnes AI?

Henry Suryawirawan: So Bruce, I think it’s been a great conversation. Is there anything else that you wanna mention about Agnes, maybe cool use cases from, you know, your journey so far that people have in terms of using Agnes? Yeah, it could be anything else as well.

Bruce Yang: Yeah, we have received emails from a lot of our users talking about how they use our product. Like we have received emails from teachers from Abu Dhabi who have mentioned that they have transitioned from Gamma to us because we are making equally good slides with much more quota and a much cheaper price. They are asking for more quota. And we have also received, you know, emails from Philippines with emails end with gov.ph, which is probably from the government sector, asking us to provide more quota for deep research because they’re doing a lot of deep research relying on AI.

So we have seen a lot of people from emerging countries adopting AI from our platforms. And one of the things which we definitely observe is they’re not afraid of AI. They’re willing to try and they’re willing to try with more advanced features. But the problem that with all the other product is they have a barrier of a paid tier, which if you don’t spend money, you only try with very basic conversations. Like if you don’t, if you’re not subscribed to ChatGPT, you won’t have deep research. And that deep research is only 15, the quota is only 15 for the entire month. For us, our research is we’re giving like close to 30 daily research for everybody for free.

And I definitely see that this is very important effort for us to get people from emerging countries to adopt AI. And we want to have the AI priority in the world and the willingness to pay should not be a barrier. And if people want to pay, you can, we provide some of the features like, I mean some of the options with starter subscription like $3 a month if people from emerging countries want to try our product with even higher quota. Even if they don’t want to try with a high quota, we provide something like, one time, you know, one time credits for them to complete the slides if they already finished half of the slides and wanna finish the second half. But basically the quota is quite a lot for everyday usage.

And we also believe the one thing which we believe is that the traffic itself is of value. So just like how TikTok is able to gain a lot of users using their product very quickly every day with a huge amount of hours, very high retention. So one thing which we, one of our measures to help, you know, adopters to try our advanced features without paying is to bring more other friends. All the family friends using a product and, you know, form a group chat so that they can talk, they can keep it, keep everybody active on the platform because we believe that the high usage, the habits, the high traffic would be even more valuable than the AR, than the paying customers. So that’s some strategy and some philosophy of us building Agnes AI. That’s something I want to share right now with you guys.

[00:57:30] How Does Agnes AI Make Money While Offering Free Features?

Henry Suryawirawan: Yeah. So out of curiosity, because you seem to be very generous providing in higher quota, not even paying for people to leverage some of the cool features. How is your business model like? Or are you focusing a lot more on acquisition at the moment? And even like we mentioned about the traffic, right? Are you leveraging on the data that comes in for further training and all that? So tell us a little bit more about this aspect.

Bruce Yang: Yeah. That’s a very good question. So we definitely focus a lot of of user acquisition this stage because we are late comer. Not like one year ago or two years ago when, three years ago, when ChatGPT came up. So we want to catch up with their traffic. That’s why we do not focus a lot on the AR, on the payment right now. But the reason that we’re able to do that is we’re able to support the high usage with our own models with very low cost. We’re like 20th, 1/20th of that of our competitors of other place.

And how we wanna make money, twofold. One still on subscription, but this is on the high paying markets like US, Japan, Korea, Europe. For this regions, we want to enter, but we will give a very big discount in the subscription. Provide equal or more features, better quota, but the cost will be one third of that of the other place. The other thought is the emerging countries. We understand that most people are not paying even, not only limited to our products, but also ChatGPT, Gemini, nobody wants to pay, especially if it is a monthly subscription. One of the reasons is you, everybody is afraid that they’re missing about, I mean, they miss the time of cancelling or unsubscribe but the product is not used anymore. So, so we try to solve the problem by giving a lot of one time payment, which is very low cost like $1 or half a dollar for people to complete a task. If you don’t really want to spend any money, any dollar, you can also get people to our platform by invite more people, you know, be very active on the group chat because the user acquisition is also cost from our side. By saving cost is making money for us.

And long-term speaking, we’re definitely, you know, believing thesis of traffic. Just like how Google, Meta, TikTok becomes so valuable because everybody is talking about it. Everybody believe about the brand. Because of brand, because of high traffic, they are able to monetize with other kind of matters like ads, like IP, like e-commerce. ChatGPT is trying similar things with the apps and the GPT, with instant checkout. We could potentially do the same thing. But we form our own ecosystem, because we have higher traffic on the emerging countries. But our business model definitely will be very different from all the other players because of our own low cost, because our selection of the market, because we, we are originated from a different place of the world. Yeah.

Henry Suryawirawan: Yeah. Definitely, you know, for those listeners who come from Southeast Asia, we can be proud of having like a model that comes originally from, you know, Southeast Asia. Singapore, specifically, right? Because like the world has seen ChatGPT. The world was disrupted by DeepSeek, you know, the open source model from China. And I hope one day I can see Agnes also making a noise in the world, you know, all these AI models. And definitely thanks for sharing those story with us today. And I will try my best to support Agnes AI.

[01:01:19] 3 Tech Lead Wisdom

Henry Suryawirawan: So as we reach the end of our conversation, I have one tradition that I always ask my guests, which is a question I call the three technical leadership wisdom. So you can think of it just like advice that you wanna give to the listeners. Any kind of wisdom or advice that you wanna give the listeners today?

Bruce Yang: Let me go one by one. Number one I think is, focus on one metric at one time. I think this is very important to us because a people can’t really stand out at the beginning. And for startups like us, we need to survive. We have to show one thing very strong before we can prove anything else. So I think we focus on DAU a lot that makes us stand out. We are like one of the fastest growing product in the region. So this is number one. Focus one metric at one time.

Number two, I think, rely a lot on your team. I think we, I rely a lot on my team, my advisor from NUS, my schoolmate from my alumni, Berkeley Alumni, who is also MIT alumni doing our research. Because I can only focus on one thing and if I focus on too many things, as a team, I can’t do well for anything. So rely a lot on your team. Find a good team so you can do something big together.

The third part about technical, I think let me think about that. I would say learn from your mistakes. You know, in my process of building my startup, I keep reflecting on myself and keep-asking whether I could, whether I did wrong and whether I could do better. And this become a ritual for me to keep changing myself, evolving myself. It’s something, you know, which I think is the beauty of doing a startup because if you don’t have the kind of pressure, if you don’t have the kind of ambition, people do not tend to evolve or change by themselves. By doing a startup with so much opportunity and such high pressure of handling a lot of things at one time, I just become, you know, a better person myself. I just be more disciplined. I just try to find my own problems and try to better myself. That’s one thing I definitely feel very valuable to everybody doing tech, technical, especially doing startups. Yeah, keeping evolving, yeah.

Henry Suryawirawan: Wow, thank you for such a beauty wisdom. So Bruce, if people love this conversation, they wanna connect with you, they wanna find out, reach out to you online, is there a place where they can find you?

Bruce Yang: Of course, I can give my email, which is bruce at sapiens dash ai.io. I can send out to you. Yeah, you can maybe put on the comment.

Henry Suryawirawan: Yeah, sure.

Bruce Yang: Or they can also find me on LinkedIn. Bruce Yang, yeah, Agnes AI.

Henry Suryawirawan: Okay, cool. So thank you so much for sharing today. So I wish Agnes AI, you know, a much greater success, a more prominent, you know, usage in the world, right? Not just Southeast Asia, definitely. And I think we all can root for the success of Agnes AI simply because, yeah, it’s one of the local models that came from Southeast Asia. So thanks again, Bruce, for sharing today.

Bruce Yang: Thank you very much, Henry. It’s very pleasure talking to you. – End –