28. OBI Tech Talk: Event Sourcing for AI & Decentralized Architecture

Join us for our second OBI Tech Talk in October!

⏰ Event-Details:
Doors open: from 6pm

18:30 PM - “Event Sourcing for AI Agents: Designing harnesses that survive Production”
Bruno Bonando - Director of Technology @ Intellias
19:15 PM - “Centralised business logic in decentralized architecture”
Michaël Le Barbier - Data & AI Engineer @ Thoughtworks
20:00 - Open discussion, casual networking & pizza!

Location: OBI Office, Carlswerk, Cologne
Talk Language: English 🇬🇧
Cost: Free admission.

Talk 1: Event Sourcing for AI Agents
Designing harnesses that survive Production
AI agents that work in a demo often struggle in production. Retries, partial failures, changing tool outputs and audit requirements quickly expose the limits of a simple prompt-and-loop design.
This session shows how event sourcing gives an agent harness a solid foundation. Every decision, tool call, and state change becomes an immutable event, so agent runs can be replayed, debugged, audited, and recovered after failure. We walk through the architecture and look at code samples for the core building blocks: the event log, projections, idempotent tool execution, and replay-based debugging and evaluation. You leave with a concrete pattern you can apply to your own agent systems.

Bruno Bonando (Director of Technology @ Intellias)
Bruno Bonando is Director of Technology at Intellias, where he leads the Retail practice for the DACH region. With more than 23 years of engineering leadership, he helps companies move AI from pilot to production across engineering, business operations, and product, with particular depth in retail and commerce. He still writes code and runs production systems himself.

Talk 2: Centralised business logic in decentralized architecture
In a monolithic or centralized architecture, a piece of business logic can be owned by a specific component and present an artefact that software engineers and product owners can easily reason about to explain or understand the behavior of the software.
In a microservice or decentralized architecture, such a code artefact vanishes as the business logic is diluted in the service mesh. Traces can recreate a tangible artefact displaying the logic result on a specific run, but they do not carry the same explainability weight as an actual procedure capturing the business logic, which makes it very difficult to reason about modifications.
In this talk, Michaël illustrates the problem with a lab example from retail, explains why monads show up naturally in a decentralized setup but have poor engineering properties, and shows how to extend the compiler to get the best of both worlds (example in Common Lisp).

Michaël Le Barbier (Data & AI Engineer @ Thoughtworks)
Michaël Le Barbier is a Data & AI Engineer at Thoughtworks, combining his strong background as a professional mathematician, programmer, and data specialist with a Math PhD, an Agrégation de Mathématiques, and Max Planck Alumni status. He specializes in software architecture, decentralized systems, and data engineering, bridging the gap between high-level architectural design and practical implementation using clean abstractions and robust engineering practices.

Parking: You can park directly in front of our main entrance (please do not park near the E-Werk). Please find a detailed map with coordinates here.
Exit in the evening: As some gates are closed in the evening, please use the gate near Teltec Köln, Schanzenstraße 29, to leave the premises.

Disclaimer: By attending, you agree that photos and videos taken during the event may be used on social media platforms such as LinkedIn to promote future events.