Agnes AI: This is How You Win Southeast Asia

Preview clips featuring key insights from the episode about Agnes AI's meteoric rise: reaching 3 million users in two months, achieving a $100 million valuation, customer acquisition cost of just 20 cents, and Bruce's vision of Agnes as the next Facebook—an AI-native social network.

Derek introduces Bruce Yang, Founder and CEO of Agnes AI, and discusses the platform's $100 million valuation. Bruce explains Agnes as a sovereign AI company from Singapore, its evolution from productivity tool to social AI platform, and the impressive growth to 3 million users in two months.

Bruce shares insights on Agnes's remarkable user growth, including a customer acquisition cost of just 20 cents, week 8 retention of 30%, and day 30 retention of 10%—comparable to Facebook's early days. He discusses what reaching 3 million users means for the platform.

Discussion on Agnes being labeled as Singapore's local homegrown AI rising star and the strategic positioning to become the king of AI products in Southeast Asia, similar to how Shopee and Lazada dominated e-commerce in the region.

Bruce explains why Agnes chose a multi-agent system over a general-purpose AI agent, discussing the evolution of AI trends from general agents to coworking environments to multimodal, and positioning Agnes as a vital lifestyle app blending productivity and social networking.

Bruce addresses comparisons with Manus, highlighting key differences: Agnes has proprietary models, focuses on mobile-first approach, targets young audiences in high-population regions, and emphasizes cultural and language support for Southeast Asia.

Deep dive into Agnes R1, a proprietary model trained using DSPO (Dynamic Phase Sequential Policy Algorithm), an enhanced version of GRPO from DeepSeek. Bruce explains the training process involving distillation from SOTA models, pre-training on user data, and reinforcement learning optimization.

Discussion on Agnes R1 achieving SOTA results for 7B models on open benchmarks like HAPAQA, and the challenge of transitioning from benchmark results to real-world problems. Bruce explains the universal verifier system for evaluating subjective tasks.

Bruce shares the journey of building Agnes's user base, starting with influencer partnerships in Southeast Asia, the magic moment when launching the mobile app, and key growth sources like ZALO in Vietnam. He discusses the strategy of rapid feature iteration to retain users.

Discussion on Agnes's success in Argentina and LATAM regions, achieving number one or two positions in several countries. Bruce explains the decision to focus on Southeast Asia first before aggressively expanding, learning from retention challenges in new regions.

Bruce explains why Agnes offers premium features for free or with generous credits, emphasizing that the North Star metric is DAU (Daily Active Users) rather than early monetization. He discusses the strategic decision to build trust and product-market fit before charging users.

Bruce shares that slide generation was an early favorite feature, but group chat with memory is becoming the most loved feature. He explains how memory creates stickiness and why proprietary memory systems are a key research focus for Agnes.

Discussion on Agnes's 32,700 token context window and the challenges of extending it. Bruce explains the use of RAG, vector DB, and file systems to manage content limits, and the advanced edit feature that solves context problems without lengthening the window.

Bruce explains Agnes's approach to token efficiency, including code agents that use pseudocode instead of natural language for multi-agent communication. This approach improves accuracy by 5-10% while reducing token usage by 40-70%, and enables fast mobile responses within one minute.

Discussion on Agnes's multimodal capabilities, including in-house image and video generation models. Bruce explains the three-agent system: prompting agent (understands human intent), generating agent (creates images/videos), and evaluating agent (provides feedback for regeneration).

Bruce shares how the collaboration with Agora started for group chat features, praising Agora's speed, cost efficiency, and fast response times. He explains how Agora's philosophy aligns with Agnes's goal to be the fastest AI consumer product mobile in the world.

Bruce shares his perspective on voice-based interaction as the future, explaining how his 9-year-old daughter uses voice exclusively for social communication. He discusses the potential for voice-first AI-native social networks and the transformative era ahead, envisioning Agnes as the next Facebook—an AI-native Facebook.

Bruce discusses the transition from general sovereign AI to social AI, positioning Agnes to compete in the social network space rather than productivity. He previews upcoming features including newsfeed, hardware integration, and community features coming in the next quarter.

Bruce encourages developers and entrepreneurs to start early in building next-gen AI products, emphasizing that the wave is bigger than imagined. He invites smart minds to join together to build an ecosystem in Southeast Asia and dominate emerging markets.

Derek thanks Bruce for joining the podcast and wishes Agnes continued success. Closing remarks and invitation to tune in for the next episode.