A few weeks ago, both my wife, Valentina, and I got invited to speak at this year's LambdaConf again. We joined last year's conference without ever hearing about it before, but it ended up being one of the best conferences we ever attended: the speakers were really high-quality and the talks were diverse and thought-provoking, the community was absolutely lovely, and the location was just gorgeous.
We were really sad when we heard that this year's conference was the last one, but we are very very grateful that we could be part of it. John De Goes, Cristina Vetencourt, and the whole team did a fantastic job in putting together an absolutely worthy final edition of LambdaConf.
The Conference
LambdaConf 2025 took place from May 12-13 in the stunning Rocky Mountains of Estes Park, Colorado, at the Holiday Inn hotel.
What made LambdaConf special wasn't just the technical content—it was the unique fusion of cutting-edge research and practical industry applications. The conference featured five parallel tracks covering functional programming, artificial intelligence, cloud-native development, and DevOps. With over 50 speakers and a diverse range of topics, it truly lived up to its reputation as a multi-disciplinary developer conference.
The conference atmosphere was warm and inviting, with thoughtfully organized activities like craft beer tasting and hiking in Rocky Mountain National Park. It was one of those conferences where there were just too many great talks at the same time, where I actually learned something from the speakers, and where I made a ton of new friends.
My Talk
My talk this year was about AI agents, and more specifically how we as developers can build and integrate agents that help us do our jobs better. The talk was titled "Cut the Hype and Build Your Own AI Agents," and I was excited to share some practical insights about moving beyond the expensive, slow, and often disappointing big-name tools like Devin.
The core message was simple: everyone keeps calling 2025 "The Year of AI Agents," but the real value comes from building custom, specialized agents for tight-scoped, high-impact tasks. Instead of relying on one-size-fits-all solutions, I showed the audience how to create agents that handle the repetitive parts of development that bog teams down when they're trying to move fast.
I walked through the fundamentals of building these specialized agents, covering how to scope them properly, integrate them into existing codebases, make them available to Cursor and other tools with MCP, and ensure they deliver actual value rather than just impressive demos.
The source code of the examples I showed is available at MaxGfeller/ai-agents-for-developers-examples.
The reception was fantastic—I had great conversations with attendees afterward about their specific use cases and challenges.
Other Interesting Talks
The conference was packed with fascinating talks that showcased the breadth and depth of modern programming language research and practice. Here are some of the standout presentations that caught my attention:
Jai' Demo and Design Explanation by Jonathan Blow was one of the keynote highlights. Blow demonstrated features of his Jai programming language and explained the reality-based design decisions behind it. Coming from his experience developing complex video games, Jai addresses specific problems that arise when building very complex software, with a focus on practical solutions rather than theoretical elegance.
Temporal Modal Logic: The Time is Now by Foy Savas offered a fascinating dive into the history and impact of temporal modal logics. From teaching computers to think to verifying network protocols, this talk explored why some programming languages struggle so much with mathematical reasoning and how temporal logic provides better tools for handling time-based systems.
What's Next for WebAssembly on the Server—and How to See It in Action by Harry Kimpel explored WebAssembly's potential beyond the browser. With new capabilities and tools emerging, WebAssembly is positioning itself as a game-changer for cloud-native applications and microservices, offering promising solutions for performance and portability challenges.
Introduction to MoonBit and its Async Model by Hongbo Zhang introduced this modern programming language taking significant steps forward with asynchronous programming. MoonBit's new async features are designed to simplify the development of scalable and efficient applications, particularly in microservices architectures.
AI'll Be Back: Exploring Diffusion Models for Image, Audio, and Video by Martin Förtsch and Thomas Endres provided an excellent introduction to generative AI across multiple modalities. They explained how neural networks can generate images, music, and short videos from text inputs using diffusion models and transformer architectures—timely given the rapid advances in this space.
Building a Compiler and Processor for Fast Processing of Encrypted Data by Ryan Orendorff presented Sunscreen's novel approach to fully homomorphic encryption (FHE). Their system allows developers to write regular C code that gets transformed into performant, secure FHE applications capable of processing encrypted data without ever decrypting it—a breakthrough for privacy-preserving computation.
The Rise of Reasoning Models by Valentina Halasi (my wife!) explored the evolution from traditional LLMs to reasoning models like OpenAI's o1 and DeepSeek R1. These models excel in step-by-step problem-solving and logical reasoning, representing a significant shift in AI capabilities toward more structured thinking and planning.
All of the talks are now available on YouTube. Make sure to check them out!
