Matouš Havlena: A Real-World Blueprint for Scalable AI Agents
How do we make AI work not just in labs but inside enterprise workflows? Matouš Havlena, co-founder and Head of Engineering at Apoco, answered that at Innovations United 2025 with clarity, humility, and powerful case studies.
“Companies don’t come to us with blueprints. They come with complex problems—and that’s where we thrive.”
Founded in 2019 with his identical twin brother, Apoco began as a family business in the Czech Republic after a stint in the U.S. working in applied research. Today, Apoco builds real AI systems for some of the world’s most demanding organizations like IBM Research in the USA.
From Robots to Satellites: Building the Future, One System at a Time
Apoco’s projects include:
- 🦾 Robotic arms orchestrated via voice and vision
- 🧪 A data processing platform for autonomous chemical labs
- 🧠 An enterprise-grade LLM platform with inference and multi-modal capabilities
Their work even touches data from the Hubble telescope—a sign of how far-reaching their innovation has become.
The Three Challenges Holding Back Enterprise AI
Matouš described three systemic barriers to AI agent adoption:
- Fragmentation – Agents built in incompatible frameworks/languages can’t talk to each other.
- Composition – It’s hard to plug small agents into workflows that actually solve problems.
- Scaling – Local agents on laptops aren’t shareable, observable, or secure enough for enterprise use.
“The next wave of AI will be made of small, composable agents that do one thing well—and delegate the rest.”
The Solution: Agent Communication Protocol (ACP)
To bridge these gaps, IBM Research launched a project to define ACP — a protocol designed to enable agents to describe and communicate across frameworks, companies, and servers. The project is technically led and developed by Apoco under contract with IBM Research, as part of their broader effort to standardize multi-agent communication.It works seamlessly alongside Google’s A2A and Anthropic’s MCP—and IBM is already collaborating with Google to align efforts into one global standard.
A Platform to Scale It: The BeeAI Framework
To tackle composition and scaling, IBM Research initiated the open-source BeeAI platform (built on ACP) with Apoco leading its development. The platform enables teams to:
- Discover and compose agents via GUI or CLI
- Deploy locally, on-prem, or in the cloud
- Integrate into real enterprise systems (auth, observability, identity)
And it’s all open source under Linux Foundation governance, contributed by IBM Research.
“AI isn’t about building smarter chatbots. It’s about enabling ecosystems of smart tools to collaborate.”
Final Word
Matouš Havlena’s keynote was a masterclass in engineering-led innovation—grounded, global, and built for the real world. Apoco isn’t just riding on the wave of AI hype, but with their partnerships Apoco focuses on building infrastructure for the next era of enterprise AI that will further unlock the potential of AI in corporations.
