Matthias Luebken is the co-founder of TAVON.ai, where he builds AI agents that work inside the email, CRM, ERP, and sales processes businesses already use. His approach draws on a career spanning agile software development, cloud-native infrastructure, developer platforms, and product leadership: give automated systems useful tools and business context while keeping consequential decisions under human control.
After studying applied computer science at Bonn-Rhein-Sieg University of Applied Sciences, Luebken worked as a developer, trainer, and agile coach at it-agile and moved into technical product management and software-development leadership at Adcloud. He subsequently led product at Giant Swarm, worked on developer tooling at Red Hat, and focused on infrastructure, cloud, and Kubernetes as a senior product manager at Instana.
His container-patterns repository documents reusable approaches to building applications with containers. In 2021, he joined Upbound as a product manager working on Crossplane and its ecosystem. His writing on developer platforms advocated accessible cloud services, useful infrastructure abstractions, production-ready operations, and open-source governance; his Crossplane provider directory mapped the surrounding ecosystem. He later became chief product officer at multicloud company emma before co-founding TAVON.ai with Ivan Pedrazas.
How he builds useful agents
- Make tools legible to agents. Luebken argues that reliability depends on clearly defined functions, understandable outputs, explicit errors, and relevant operational context. His case against making agents guess treats interface design as essential infrastructure for dependable automation.
- Apply coding-agent infrastructure beyond coding. His a-piece-of-pi project adapts Pi’s agent core and coding-agent extensions to CRM lead qualification, with event streaming, interactive commands, and confirmation gates before customer records change. He favors small, composable command-line tools that let agents use established software without elaborate integration layers.
- Keep automation inside familiar workflows. A client sales system routes incoming requests to customer-specific agents, retrieves CRM and ERP data, maintains persistent case sessions, and prepares editable email drafts. His AI Engineer Europe talk also addresses permission checks and the unresolved challenges of sandboxing and policy enforcement.
Luebken organizes AI Tinkerers Cologne and has raised questions about AI-assisted teams, including changing management responsibilities and how junior developers learn when more work involves directing and reviewing automated systems.
