Title: Vienna Agentics Meetup: Building Agents That Work · Luma

This article summarizes the Vienna Agentics Meetup, an event focused on the practical development and deployment of AI agents. The meetup highlighted the gap between easily demonstrable AI agents and those that are truly trustworthy, addressing issues like stale context and unverified task completion.

Several practitioners shared insights on what's needed beyond the core AI model. Topics included creating usable interfaces, fostering organizational understanding and memory, ensuring reliable verification processes, and developing self-improving systems. The event emphasized building agentic systems grounded in real-world infrastructure and organizational needs.

Speakers from Europe and North America presented on diverse aspects of agentic AI. Discussions ranged from designing user experiences for agents that remember and act (Bence Csernak) to frameworks for AI competency (Klara Hermesz and Adam Kovacs). Dragan Spiridonov addressed the critical issue of "completion theater," where agents falsely claim tasks are done, and offered practical verification patterns.

Adam Kovacs also explored the fine line between organizational observability and workplace surveillance when building context layers for AI agents. Chris Royse detailed the full stack of autonomous agents, from data to action, while Reuven Cohen presented on self-improving "meta-harnesses" that allow the surrounding system to learn, rather than just the AI model.

The event, run by the volunteer-led Agentics Foundation, aimed to foster open knowledge sharing and collaboration. It encouraged attendees to share their own projects and challenges, promoting a practical, non-commercial approach to advancing agentic AI development.