Zeta Alpha - Improving Multi-Agent Workflows with Self-Optimization

​As part of Amsterdam Tech Week 2026, join Arthur Câmara (Senior Research Engineer @ Zeta Alpha), co-author of Self‑Optimizing Multi‑Agent Systems for Deep Research, for an interactive talk/discussion on improving multi-agent workflows for industrial applications.

​We’ll look at how to allow multi-agent systems to explore their own traces, reason about previous failures and continue improving with self-feedback. We will also discuss how other ideas for automatically refining an agent harnesses, like autoresearch may (or may not) work in practice, and how we can help agents to trully learn with past failures. Bring your technical questions for the Q&A, and stay for a drink afterwards!

​Who it’s for: Engineers, product/tech leads and builders working on agentic systems, LLM applications, and AI engineering/dev tools.

​What we’ll cover:

• ​Different approaches to allow multi-agent systems to explore and refine their own harnesses.

• ​How methods like GEPA and TextGrad allow agents to refine its own prompts over time, and how much improvement we can expect.

• ​Industrial application: How much does it cost? What if you have no data to get started? How about auto-research?

​

​See the full AMS Tech Week calendar.