←
Bruce Yang213 posts
Bruce Yang avatar

Bruce Yang

@Bruce_why

Founder of Sapiens (@agnesai_sapiens) | Turing Award Disciple | Building Frontier AI Models & Harnesses

◷ Joined June 2011▧ 135 Photos and videos
Bruce Yang avatar
Bruce Yang@Bruce_why·Nov 17

What happens after an AI system gets something wrong matters to me. Can it learn from that attempt and do better next time, without needing a full model retrain? At Agnes AI, we’re exploring two approaches, now featured in Tech in Asia. With RSI, an AI coach reviews previous games and improves the team’s playbook while the underlying models stay unchanged. In one Honor of Kings experiment, the win rate rose from 0% to 100% after five rounds under the test conditions. With TTT, we update selected model parameters. In a simulated robotic-arm task, success improved from about 30% to around 90%. These experiments test different ways for AI to adapt. We’re already talking with companies in robotics, medical technology and other industries about applying our models and agent systems to their work. That’s where we want to take this research next. Thank you to Terence Lee and the Tech in Asia team for covering our research and sharing our progress. Read the full story: techinasia.com/news/singapor… #AgnesAI #AIResearch #ArtificialIntelligence

Bruce Yang avatar
Bruce Yang@Bruce_why·Nov 11

For years, AI progress has largely been driven by scaling models, data, and compute. But deploying AI in the real world presents a different challenge. AI cannot spend months retraining every time it encounters a new task or environment. At Agnes AI, we're working to change that. As featured in TheStraitsTimes, our approach has reduced AI training time for robotics from months to as little as 30 minutes in our experiments, by training only a fraction of a model's parameters. But speed is only part of the equation. I believe the future lies in AI systems that continuously learn from experience, adapt to new environments, and operate reliably across industries, from robotics and healthcare to enterprise operations. 𝙊𝙪𝙧 𝙜𝙤𝙖𝙡 𝙞𝙨𝙣'𝙩 𝙨𝙞𝙢𝙥𝙡𝙮 𝙩𝙤 𝙢𝙖𝙠𝙚 𝘼𝙄 𝙢𝙤𝙧𝙚 𝙥𝙤𝙬𝙚𝙧𝙛𝙪𝙡. 𝙄𝙩'𝙨 𝙩𝙤 𝙢𝙖𝙠𝙚 𝙞𝙣𝙩𝙚𝙡𝙡𝙞𝙜𝙚𝙣𝙘𝙚 𝙖𝙙𝙖𝙥𝙩𝙖𝙗𝙡𝙚, 𝙘𝙤𝙨𝙩-𝙚𝙛𝙛𝙞𝙘𝙞𝙚𝙣𝙩, 𝙖𝙣𝙙 𝙥𝙧𝙖𝙘𝙩𝙞𝙘𝙖𝙡 𝙛𝙤𝙧 𝙧𝙚𝙖𝙡-𝙬𝙤𝙧𝙡𝙙 𝙙𝙚𝙥𝙡𝙤𝙮𝙢𝙚𝙣𝙩. Thank you to KennyChee and TheStraitsTimes for covering our work and vision. Read the full story: straitstimes.com/tech/singap… #ArtificialIntelligence #AgnesAI #AIResearch

Bruce Yang avatar
Bruce Yang@Bruce_why·Nov 8

𝐓𝐡𝐞 𝐅𝐮𝐭𝐮𝐫𝐞 𝐨𝐟 𝐀𝐈: 𝐖𝐡𝐚𝐭'𝐬 𝐍𝐞𝐱𝐭? Great to join the "Investing in the Era of AI" panel at B71's AI Accelerate 2026. I enjoyed exchanging perspectives with fellow founders and investors on the future of AI and entrepreneurship. I believe strong AI models are the foundation of a successful AI company. But long-term success also depends on turning technical capabilities into products customers genuinely need, building differentiated technology, and creating a sustainable business. At Agnes AI, we're advancing our AI models while developing Agnes Harness, an open-source execution framework that connects AI models with tools and real-world systems. This is how we're working toward our vision of "𝐎𝐧𝐞 𝐁𝐫𝐚𝐢𝐧 𝐟𝐨𝐫 𝐄𝐯𝐞𝐫𝐲𝐭𝐡𝐢𝐧𝐠." Thanks to 𝐁𝟕𝟏, 𝐌𝐢𝐜𝐫𝐨𝐬𝐨𝐟𝐭, 𝐚𝐧𝐝 𝐄𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞 𝐒𝐢𝐧𝐠𝐚𝐩𝐨𝐫𝐞 for bringing the community together! Explore Agnes AI: agnes.ai/ #AgnesAI #AIAccelerate #ArtificialIntelligence #AIInnovation #Entrepreneurship

Post attachment
Bruce Yang avatar
Bruce Yang@Bruce_why·Oct 28

AI should do more than answer questions. It should be able to act, learn, and improve. At Agnes AI, we're building an integrated AGI system connecting reasoning, task execution, physical-world interaction, and learning from experience. Professor Xiaofan Li and I discuss how Agnes Harness and MHS contribute to this vision, and what comes next.

3:40
Bruce Yang avatar
Bruce Yang@Bruce_why·Oct 3

𝐖𝐡𝐲 𝐈 𝐂𝐡𝐨𝐬𝐞 𝐄𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐜𝐲 𝐎𝐯𝐞𝐫 𝐁𝐢𝐠𝐠𝐞𝐫 𝐁𝐮𝐝𝐠𝐞𝐭𝐬 When we started building Agnes AI, I didn't want our advantage to come from simply spending more money on training. I wanted to prove that a globally competitive AI model could be built with a fundamentally more efficient approach. That philosophy now extends across text, image, and video. By pushing down both training and inference costs, we can make powerful AI much more accessible and, in some cases, free at scale. I believe the next generation of AI companies will not be defined only by how capable their models are, but by how efficiently they can deliver that intelligence to the world. #ArtificialIntelligence #AIModels #GenerativeAI #AgnesAI

1:53
Bruce Yang avatar
Bruce Yang@Bruce_why·Sep 28

𝐖𝐡𝐲 𝐈 𝐁𝐞𝐥𝐢𝐞𝐯𝐞 𝐁𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤𝐬 𝐌𝐚𝐭𝐭𝐞𝐫 𝐟𝐨𝐫 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐓𝐫𝐮𝐬𝐭 𝐢𝐧 𝐀𝐈 I believe benchmarks matter because they give us an objective way to understand whether our models are actually improving. Without consistent evaluation, iteration becomes slower and much more subjective. At Agnes, we treat globally recognized benchmarks as part of the development process, not just something to highlight after launch. Our text, image, and video models have all reached Top 10 positions in global rankings. For me, that external validation matters because trust should be built on measurable performance, especially when working with enterprises and governments.

1:16
Bruce Yang avatar
Bruce Yang@Bruce_why·Sep 19

𝐖𝐡𝐚𝐭 𝐈 𝐒𝐡𝐚𝐫𝐞𝐝 𝐚𝐭 𝐓𝐞𝐜𝐡 𝐢𝐧 𝐀𝐬𝐢𝐚 𝐂𝐨𝐧𝐟𝐞𝐫𝐞𝐧𝐜𝐞 I had the chance to share what we're building at Agnes AI today — from our own text, image, and video models to APIs, AgnesCode, and the product layer built on top of them. What matters to me is not just building stronger models, but turning them into a production-ready stack that developers and companies can actually use. Great to meet so many builders, founders, and teams at Tech in Asia Conference, and to hear the questions and feedback directly from the community. #AgnesAI #TechInAsia #ArtificialIntelligence #AIModels #DeveloperTools

Post attachmentPost attachmentPost attachmentPost attachment
Bruce Yang avatar
Bruce Yang@Bruce_why·Sep 10

Proud to see Sapiens Technology and Agnes AI featured by Analytics Insight following our inclusion in the Forbes Asia 100 to Watch 2026. Recognition is encouraging, but independent benchmarking matters even more to me. It is where claims meet measurable results, and where developers can judge for themselves how competitive a model really is. As Agnes models enter more third-party benchmarks, our focus remains the same: build capable AI that developers can actually use, afford, and bring into real products. analyticsinsight.net/artific…

Post attachment
Bruce Yang avatar
Bruce Yang@Bruce_why·Sep 7

I like seeing AI coding move beyond static pages. This 3D bookshelf demo, built with Agnes 2.5 Flash on AgnesCode, is a good example of how interaction, motion, 3D design, and product thinking can come together in one build. For me, this is where AI coding gets more interesting — not just generating code, but helping turn ideas into experiences people can actually use and explore. agnes-ai.com/

𝐂𝐚𝐧 𝐀𝐈 𝐁𝐮𝐢𝐥𝐝 𝐚 𝟑𝐃 𝐈𝐧𝐭𝐞𝐫𝐚𝐜𝐭𝐢𝐯𝐞 𝐁𝐨𝐨𝐤𝐬𝐡𝐞𝐥𝐟? Built with 𝐀𝐠𝐧𝐞𝐬 𝟐.𝟓 𝐅𝐥𝐚𝐬𝐡 on 𝐀𝐠𝐧𝐞𝐬𝐂𝐨𝐝𝐞. This demo turns a simple bookshelf concept into an interactive 3D web experience with animated books, hover states, detailed pages, and smooth transitions. A glimpse of what AI coding can build beyond static interfaces.

0:30
Bruce Yang avatar
Bruce Yang@Bruce_why·Sep 4

𝐖𝐡𝐲 𝐃𝐢𝐝 𝐖𝐞 𝐃𝐞𝐜𝐢𝐝𝐞 𝐭𝐨 𝐓𝐫𝐚𝐢𝐧 𝐎𝐮𝐫 𝐎𝐰𝐧 𝐌𝐨𝐝𝐞𝐥𝐬? I didn't start Agnes with the ambition of building our own foundation models. In the beginning, we trained models mainly to support our own applications. But as our user base grew into the millions, relying entirely on external APIs quickly became difficult to justify at scale. That pushed us to build more of the stack ourselves. By the middle of last year, our own models were already handling around 80–90% of our internal model usage. As performance improved across text, image, and video — together with much stronger cost efficiency — we realized the same infrastructure we built for ourselves could also create value for other companies. Today, that is becoming an increasingly important part of Agnes: training our own models and making them available through APIs so developers and businesses can access capable AI at a much more practical cost. agnes.ai/ #ArtificialIntelligence #AIModels #GenerativeAI #AIInfrastructure #AgnesAI

2:00
Bruce Yang avatar
Bruce Yang@Bruce_why·Sep 2

I Believe Frontier AI Shouldn’t Require a Frontier Budget Agnes 2.5 Pro Alpha is now featured on VentureBeat following its debut on Artificial Analysis. What matters most to me is not just strong performance, but making capable AI affordable enough for more developers and teams to actually use at scale. That’s a big part of what we’re building toward at Agnes. venturebeat.com/business/fro…

Bruce Yang avatar
Bruce Yang@Bruce_why·Aug 26

Proud to see Sapiens Technology named to the 𝐅𝐨𝐫𝐛𝐞𝐬 𝐀𝐬𝐢𝐚 𝟏𝟎𝟎 𝐓𝐨 𝐖𝐚𝐭𝐜𝐡 𝟐𝟎𝟐𝟔. From the beginning, we believed that a focused team could train its own AI models, build across modalities, and compete globally by being faster, more efficient, and relentlessly focused on real-world usefulness. That belief is still at the core of how we are building Agnes today. Recognition like this is encouraging, but for us, it is also a reminder that we are still early. There is much more to build. Thank you Forbes Asia, and thank you to everyone on the team who made this possible. forbes.com/sites/ranawehbe/2… #ForbesAsia #ArtificialIntelligence #AIInnovation #AgnesAI

Post attachmentPost attachment
Bruce Yang avatar
Bruce Yang@Bruce_why·Aug 22

𝐃𝐨 𝐅𝐫𝐨𝐧𝐭𝐢𝐞𝐫 𝐀𝐈 𝐂𝐨𝐦𝐩𝐚𝐧𝐢𝐞𝐬 𝐇𝐚𝐯𝐞 𝐭𝐨 𝐂𝐨𝐦𝐞 𝐅𝐫𝐨𝐦 𝐭𝐡𝐞 𝐔𝐒 𝐨𝐫 𝐂𝐡𝐢𝐧𝐚? For the past few years, it was easy to think so. The strongest concentration of AI talent, capital, and GPU resources was largely in a few major technology hubs. But that landscape is starting to change. Talent is becoming more globally accessible, compute infrastructure is expanding, and the open-source community is playing a much bigger role in how AI progresses. When we started Agnes, this shift gave us the confidence that a globally competitive AI company could also be built from Singapore, while drawing on talent and resources across the wider region. I believe the next generation of frontier AI companies will not be defined only by where they are based, but by how effectively they can bring together talent, infrastructure, and ideas from around the world. agnes-ai.com/ #ArtificialIntelligence #FutureOfAI #AIInnovation #AgnesAI

1:04
Bruce Yang avatar
Bruce Yang@Bruce_why·Aug 19

𝐖𝐡𝐞𝐫𝐞 𝐖𝐢𝐥𝐥 𝐀𝐈 𝐕𝐚𝐥𝐮𝐞 𝐌𝐨𝐯𝐞 𝐍𝐞𝐱𝐭? Foundation models are improving rapidly, and the gap between leading models is narrowing faster than many expected. As that happens, differentiation will increasingly move beyond the base model itself. Post-training, orchestration, deployment, and real business outcomes will become increasingly important. At Agnes, we are paying close attention to this shift. Building strong models still matters, but the next phase of AI will also be defined by how well these models can work together and create real-world value. 🔗 agnes-ai.com/ #ArtificialIntelligence #AIModels #AIInnovation #GenerativeAI #AgnesAI

Post attachment
Bruce Yang avatar
Bruce Yang@Bruce_why·Aug 16

𝐂𝐚𝐧 𝐀𝐈 𝐋𝐞𝐚𝐫𝐧 𝐇𝐮𝐦𝐚𝐧 𝐕𝐚𝐥𝐮𝐞𝐬? When we raise an animal, we teach it how to behave around humans. AI is different — its understanding of us comes from something much larger: the knowledge, ideas, and values accumulated by humanity over generations. AI is not learning from one person, but from human civilization itself. If our collective knowledge reflects our desire for progress and collaboration, those values can also shape the intelligence we create. At Agnes, we believe building AI is not only about capability, but also about creating systems that can better collaborate with humans. agnes-ai.com/ #ArtificialIntelligence #FutureOfAI #AIInnovation #AgnesAI

1:21
Bruce Yang avatar
Bruce Yang@Bruce_why·Aug 7

𝐃𝐨 𝐰𝐞 𝐫𝐞𝐚𝐥𝐥𝐲 𝐧𝐞𝐞𝐝 𝐭𝐡𝐞 𝐦𝐨𝐬𝐭 𝐩𝐨𝐰𝐞𝐫𝐟𝐮𝐥 𝐀𝐈 𝐦𝐨𝐝𝐞𝐥 𝐟𝐨𝐫 𝐞𝐯𝐞𝐫𝐲 𝐭𝐚𝐬𝐤? The last 10% of AI capability may matter for the most complex problems, but the majority of users rely on AI for mainstream, everyday applications. For 90% of use cases, efficient models can already deliver strong results at a much lower cost. The future of AI is not about using the biggest model for everything, but choosing the right intelligence for each task. At Agnes, we are building capable and efficient AI systems to make advanced intelligence more accessible. Developers can already explore our models through our free API access. Learn more: agnes-ai.com #ArtificialIntelligence #GenerativeAI #AIModels #MachineLearning #AgnesAI

0:57