AI Developer

Chairpersons welcome and opening remarks

What’s real vs. hype: Where GenAI is already delivering value in dev workflows, and where it still falls short.

• Velocity vs. quality: Does GenAI speed up coding without introducing new risks?
• Integration into the toolchain: How devs are embedding GenAI into IDEs, CI/CD, testing, and code review
• Shaping engineering culture: How pair programming with AI is changing team collaboration and onboarding
• Risk and control: Versioning, reproducibility, hallucinations — what engineers need to watch for
• What’s next: Multi-agent dev environments, AI debugging, and the future of human-in-the-loop coding

As AI agents move into production, grounding enterprise data is no longer enough. This session explores how a Fortune 500 insurer engineered an identity-aware memory layer using entity resolution to improve agent reliability and decision-making. Covering architecture, system integration, engineering trade-offs and deployment lessons, attendees will gain practical patterns for building more accurate, production-ready AI agents.

Getting an LLM to answer a question is easy. Building an AI application that remains reliable, observable and maintainable at scale is not. This session explores the emerging engineering discipline around production AI, including orchestration, context engineering, evaluation, memory, observability and governance and why these systems are becoming more like distributed software than traditional machine learning.

Building an AI agent that can query enterprise data is easy to demo, but much harder to make reliable, secure and production-ready. This session explores how Model Context Protocol (MCP), semantic metadata, deterministic guardrails and evaluation can be combined to give AI agents trusted access to enterprise data. Attendees will learn practical architectural patterns for moving agentic AI from proof-of-concept into production.

As AI applications evolve from conversational interfaces to intelligent systems capable of reasoning, planning, and taking action, development teams are rethinking how AI products are designed and deployed.

This panel brings together AI engineers, architects, and technology leaders to explore the emerging architectural patterns behind modern AI-native applications. From retrieval-augmented generation (RAG) and agentic workflows to context management, memory systems, orchestration frameworks, and evaluation pipelines, panellists will discuss the technologies and design decisions enabling more capable, reliable, and scalable AI experiences.

Join the discussion as experts share lessons from production deployments, highlight common pitfalls, and examine how application architectures are evolving to support the next generation of AI-powered products.

The faster path to production AI is an API key and a frontier model. GLS chose control instead: a customer-service system built on self-hosted, open-source small language models, live across several countries in Europe, automating millions of customer interactions between GLS and the people receiving their parcels. This session is explores the engineering road, which started two years ago – building a multi-agent solution with SLMs from scratch, while meeting regulatory requirements. Bernat Coma Puig, AI Lead at GLS, will cover the architecture design, the challenges and how to solve them: from guardrails, scalability, response accuracy, and keeping costs predictable and under control at scale.