AgentCon Singapore 2025: The 3 Big Ideas from Every Speaker
On September 9th, AgentCon Singapore brought together over 800 passionate attendees and 15 international experts for a day that redefined what’s possible with agentic AI. This event was made possible with the help of our valuable sponsors AvePoint, SUTD, MaivenPoint, and Microsoft. To kick things off on a high note, we had the pleasure of hearing directly from SUTD, who shared compelling case studies demonstrating how they’ve empowered individuals across all age groups and skill levels to harness AI for uplifting lives. These case studies span areas such as work resilience, education and training, and creativity and innovation.
This blog article is going to give you a glimpse into each session and distill the three most important lessons from the speakers. If you want to stay ahead in AI, these are the insights you need.
Dux Raymond Sy - Chief Brand Officer, AvePoint | Microsoft MVP
From Copilot to Catalyst
From the moment Dux Raymond Sy stepped onto the stage, the atmosphere was electric. Dux’s ability to weave together the history and future of AI, moving from predictive models to generative intelligence and now agentic systems, made the audience feel like they were witnessing a pivotal moment in tech. At the end of the sharing, the audience got the chance to walk away (or should I better say dance away) with the following points.
Top 3 Takeaways:
- Build a strong data foundation: AI is only as good as the information it’s built on.
- Integrate AI into daily workflows: Empower every employee to amplify productivity with AI assistance.
- Scale up with governance and trust: Ensure AI adoption is secure, compliant, and sustainable.
Dona Sarkar - Chief Troublemaker - Copilot and AI Extensibility, Microsoft
Human in the Loop: Guardrails for AI’s Wild Side
Dona’s keynote was a masterclass in curiosity and challenge. She posed the BIG questions: Would you trust AI to drive your car? Would you trust it to drive your children to school? Is AI a threat to developers, or a catalyst for new opportunities? It became clear that most people do not trust AI with truly important tasks. Her session was a whirlwind of provocative ideas and practical advice, opening attendees’ eyes to the difference between agentic AI and sparkling automation.
Top 3 Takeaways:
- Become AI users before building complex AI: Start small and improve today’s workflows.
- Humans in the loop are essential: Trustworthy AI requires human oversight.
- Put human guardrails on all inputs and outputs: Prevent exploitation and ensure reliability.
Multi‑Agent orchestration in Copilot Studio
Rene’s session offered a compelling glimpse into the future of Microsoft Copilot Studio. He showcased examples of agents designed to answer Q&A from Star Wars and Star Trek. Rene demonstrated how Copilot Studio’s low-code platform is set to democratize agentic AI, enabling anyone to create collaborative, specialized agents. His preview of multi-agent orchestration was both technically insightful and imaginative, illustrating how agents can work together, share expertise, and honor user permissions. The audience left energized, brimming with ideas about how these innovations could transform their own workflows.
Top 3 Takeaways:
- Specialized agents work together: Design agents for specific tasks and orchestrate their collaboration.
- Agent reusability: One agent can be leveraged by multiple others, boosting efficiency.
- Copilot Agents respect user permissions: Security and compliance are built-in.
Bringing AI Agents to Life with Azure AI Foundry
Henk’s session was a hands-on journey through the world of language models and agent architectures. He broke down complex concepts with clarity, then brought them to life with a live demo, an AI agent ordering pizza, complete with a human-like avatar and realistic voice. The crowd erupted when the agent refused to order pineapple pizza (It shows that even bots have opinions!). The technical workshop that followed gave attendees the tools to start building their own agents, making the abstract tangible.
Top 3 Takeaways:
- Hands-on coding accelerates learning: Use GitHub Codespaces for rapid prototyping.
- Azure AI integration is seamless: Easily connect OpenAI models with Azure services.
- Try Azure AI Agents Service: Join the Discord community and experiment with real-world builds.
NLP, LLM and AI Agents: A great combination
Huy’s session was a thoughtful exploration of the evolution of language technology. He traced the journey from foundational NLP to the rise of LLMs and now agentic AI, urging the audience to choose the right tool for each job. His insights challenged the notion that bigger is always better, showing how traditional methods can outperform LLMs for many tasks. The audience left with a deeper appreciation for efficiency and precision in AI.
Top 3 Takeaways:
- Not every task needs an LLM: Use traditional NLP for simple, specific jobs.
- Agents combine reasoning and efficiency: Leverage LLMs for planning, NLP for speed.
- Choose the right tool for the job: Optimize for accuracy, cost, and performance.
Mary Angiela Cerbolles - SharePoint / Power Platform Administrator | Microsoft MVP
Agent Activated: The New Era of SharePoint Collaboration
Mary’s session was a revelation for anyone working in digital collaboration. She demonstrated how SharePoint Agents are revolutionizing the way teams interact with content, automate tasks, and boost productivity. Her practical demos made AI accessible to every organization, showing how natural language interaction and context-aware answers can transform daily work.
Top 3 Takeaways:
- SharePoint Agent is a secure, AI-powered assistant: Works with your files, folders, and site pages.
- Natural language interaction enhances productivity: Context-aware answers save time.
- Knowledge Agent organizes content for AI-readiness: Prepares your data for tools like M365 Copilot.
Transforming Contact Centers with Scalable AI: Leveraging AI for Real-Time Voice Intelligence
Ashish’s session was a live showcase of how multimodal AI is transforming call centers and user experience. The interactive session, including a live conversation with an audience member, highlighted the power and empathy of modern AI.
Top 3 Takeaways:
- Azure AI Speech enables natural, real-time voice experiences: Unlocks new use cases in customer service and accessibility.
- Voice Live API unifies speech and generative AI: Build flexible, multilingual voice agents.
- Video Translation Service localizes content at scale: Cross language and cultural boundaries with empathy.
Building a Conversational AI Agent with Google ADK (A Hands-On POC)
Chinedu’s hands-on workshop was a deep dive into Google’s Agent Development Kit (ADK). He showed how developers can quickly build conversational agents that understand, retrieve, and summarize information, making AI practical and powerful. The session was packed with real-world examples and technical walkthroughs, inspiring attendees to experiment with ADK in their own projects.
Top 3 Takeaways:
- Fast prototyping with ADK: Build agents quickly with minimal boilerplate.
- Power of tools: Extend LLMs with external APIs and data sources for real-world impact.
- Structured and deployment-ready: ADK enforces clear separation for maintainability and flexibility.
Dmitry Katson - AI & BC MVP | Creator of CentralQ.ai, Dona Sarkar - Chief Troublemaker - Copilot and AI Extensibility, Microsoft
Multi-Agent Systems in Business
Dmitry, teaming up with Dona, brought multi-agent systems to life. Their “agent vs human” challenge was a highlight, showing how teams of agents can collaborate, compete, and extract insights from live business data. Kudos to Rene Modery who bravely took this challenge upon himself. The session made the complexities of multi-agent orchestration feel accessible and actionable.
Top 3 Takeaways:
- Start simple: Begin with a single agent and iterate.
- One agent can be enough: A well-built agent can run tools and pipelines in a loop for optimal results.
- Manager agents matter: For complex tasks, manager agents keep multi-agent systems organized and effective.
Michelle Sandford - Microsoft Developer Engagement Lead Asia
Building Trustworthy AI Agents
Michelle’s session tackled the big question: Can we trust AI with real-life decisions? Her insights on reliability, security, and human oversight were both sobering and empowering, giving the audience a roadmap for building trustworthy systems. She emphasized the need for explainability, auditability, and continuous improvement, making trust a core design principle.
Top 3 Takeaways:
- Design for reliability under uncertainty: Build failover strategies and continuous learning loops.
- Security as core design: Implement agents with least privilege and robust testing.
- Observable trust signals: Monitor, audit, and ensure human escalation paths.
Chat with your Data Using Vector Search
Pio’s session was a masterclass in Retrieval-Augmented Generation (RAG). He showed how Azure SQL and SQL Server 2025 are bringing native vector support, REST API integration, and AI-ready features to structured data, making intelligent retrieval and reasoning possible at scale. His technical walkthroughs made advanced concepts feel within reach for every attendee.
Top 3 Takeaways:
- Native vector support in Azure SQL and SQL Server 2025: Store and query embedding data directly.
- REST API integration via SQL stored procedures: Seamlessly interact with external services.
- AI-ready features in SQL Server 2025: Fast indexing, model storage, and orchestration for advanced AI applications.
Scalable AI with Memory: Orleans and Semantic Kernel for Distributed Agents
Samson’s technical deep dive into Orleans and Semantic Kernel was both advanced and practical. His demo of agent creation using a Grain showed how to build intelligent, distributed AI agents with memory, tool support, and seamless scalability. The session gave attendees a blueprint for building robust, scalable agentic systems.
Top 3 Takeaways:
- Durable memory at scale: Orleans enables agents with persistent state and identity.
- Focus on business logic: Orleans handles lifecycle and concurrency, freeing developers to innovate.
- Safe, scalable orchestration: Combine Semantic Kernel and Orleans for distributed, resilient agents.
Elian Van Cutsem - Lead DevRel @ React Bricks | CTO @ Vulpo | Astro Maintainer
Playing with Fire: Unethical AI
Elian’s session was a wake-up call about the messy realities of AI ethics. With real-world examples, some humorous and some unsettling, Elian challenged us to confront the gray areas and take responsibility for the systems we build. His practical tips made ethics feel actionable, not just theoretical.
Top 3 Takeaways:
- Ethics isn’t binary: The line between fun and harm is easy to cross; be honest about the gray areas.
- Every automation embeds a value judgment: Technical choices can have big ethical effects.
- Responsibility stays human: No matter how advanced AI gets, humans remain accountable for outcomes.
Nicholas Chang - Microsoft Azure MVP | Senior Platform Engineer@ Kainos
Supercharge Your Development with Azure MCP in Visual Studio Code
Nicholas introduced the Model Context Protocol (MCP), a universal interface for LLMs that’s set to transform agent development. His session revealed how MCP streamlines integration, boosts security, and turns VS Code into an agentic IDE. The audience left excited about the future of developer productivity.
Top 3 Takeaways:
- MCP is the USB-C for AI context: Standardizes tool and data integration across platforms.
- Accelerate development: Reduce boilerplate and reuse integrations with MCP SDKs.
- Security first: MCP follows Zero Trust and OWASP best practices for safe agent deployment.
Goloknath Mishra - Digital Transformation l Technology | Microsoft MVP
Build your first agent with Azure AI Agent Service
Goloknath lead a hands-on workshop with the following scenario. Imagine you are a sales manager at Contoso, a multinational retail company that sells outdoor equipment. You need to analyze sales data to find trends, understand customer preferences, and make informed business decisions. To help you, Contoso has developed a conversational agent that can answer questions about your sales data.
Top 3 Takeaways:
- There are many options of agent development in within the Microsoft ecosystem.
- Comparison is key: Compare the available options and choose the one that makes most sense for your use case.
- Try working on a real-life use case to deepen your understanding. For example, analyzing sales data.
AgentCon Singapore 2025 was a showcase of what’s possible when passionate people, cutting-edge technology, and practical wisdom come together. If you attended, share your favorite takeaway with us at singapore@globalai.community!