Why Your Agents Fail. Lessons from Harness Engineering
- Date
- 2026-05-02
- Location
- München, BY, Germany
- Host
- Rasa Events
About this event
Most AI agent demos look convincing right up until they meet a real workflow, real users, and real constraints. This meetup is about that gap: why agents fail in practice, what engineering teams keep getting wrong, and how to build systems that are actually reliable enough to matter. If you care about AI beyond the prototype stage, this is a chance to hear grounded lessons drawn from engineering experience, compare notes with other practitioners, and leave with a clearer mental model for what makes agent-based systems break. About the Event This is an in-person meetup in München focused on the practical realities of building and deploying AI agents. The theme is direct by design: not how to make agents sound impressive, but how to understand where they fall apart and what strong engineering looks like when you want them to perform in the real world. Expect a community-driven format that combines technical insight with room for conversation. Rather than staying at the level of hype or abstract theory, the discussion is centered on implementation lessons, tradeoffs, and patterns that emerge when teams move from experimentation into production-minded thinking. The event is also a good fit for people who learn best by talking through problems with others. Alongside the main content, there will be space to meet local builders, exchange ideas, and hear how others are approaching similar challenges across product, engineering, and applied AI work. What to Expect The core of the meetup will revolve around lessons from engineering experience: why agents fail, the assumptions that cause trouble, and the design choices that can improve outcomes. That may include topics like reliability, evaluation, orchestration, tool use, error handling, and the operational complexity that often shows up only after an agent leaves the demo environment. You should expect a format that feels useful rather than overly formal. The goal is to help attendees sharpen their thinking through concrete examples and practical discussion, not just passive listening. Likely areas of focus include: Common failure modes in agent workflows Why promising prototypes often break at the system level The engineering habits that reduce brittleness How to think about guardrails, observability, and evaluation Tradeoffs between autonomy, control, and user trust Lessons that are relevant whether you are building, testing, or integrating agents There will also be a community and networking dimension throughout the afternoon. If you have your own hard-won opinions about agents, this is the kind of room where those conversations will make sense quickly. Why Attend There is a big difference between being excited about agents and understanding how to make them dependable. This meetup is valuable because it focuses on the second problem. You will get a more honest picture of where current agent approaches succeed, where they tend to disappoint, and what experienced teams pay attention to when the stakes are higher than a demo. For technical attendees, the benefit is practical perspective. You will likely come away with better language for diagnosing failures, stronger instincts about architecture and evaluation, and a clearer sense of which shortcuts create long-term pain. For product-minded attendees, this is a chance to better understand capability boundaries. That matters when deciding what to ship, how to set expectations, and where human oversight or workflow design matters more than another layer of automation. Just as importantly, the event brings together a local community around AI, engineering, and applied systems thinking. The people in the room are likely to be asking serious questions, testing ideas in practice, and interested in conversations that go beyond surface-level excitement. Practical Details This event takes place in person in München, Germany. Being there physically matters: the format is well suited to discussion, follow-up questions, and meeting other attendees who are working through similar challenges in AI and engineering. It is scheduled for Saturday, May 2 at 2:00 PM GMT+2, making it an accessible afternoon meetup for people who want substantial content without needing to carve out an entire workday. If you are local to München or nearby, it is a strong opportunity to connect with the regional AI and builder community face to face. The tags for the event include AI, community, networking, meetup, and social, which gives a good sense of the balance: expect substance, but also an open and conversational atmosphere. You do not need to arrive with all the answers. Curiosity, practical experience, and a willingness to engage thoughtfully will go a long way. If you have been building with agents, evaluating whether they belong in your stack, or trying to separate useful patterns from noise, this meetup is designed to help you think more clearly. Come ready to listen, question assumptions, and have better conversations about what actually works.
Who should attend
This is for people who want a more realistic, engineering-grounded view of AI agents and the systems around them. - You are an engineer, developer, or technical lead working with LLM applications, agent workflows, or AI-powered products and want sharper instincts about failure modes and reliability. - You are building prototypes and starting to feel the pain of moving from something that demos well to something that needs to behave consistently in practice. - You work in product, applied AI, or technical strategy and need a clearer sense of what agents can responsibly do today versus where they still need tighter constraints and better system design. - You enjoy meetups where the conversation goes beyond hype and into architecture, tradeoffs, evaluation, and operational lessons. - You want to meet other people in München who are actively thinking about AI systems, experimentation, deployment, and the realities of making these tools useful. - You do not need to be an expert, but you will get the most from this event if you are curious, hands-on, and interested in practical discussions about what breaks and why.