Building a generative AI demo is easy. Running it reliably in production is not.

What does it actually take to move generative AI systems from experimentation into real production environments? What are the architectural, operational, and organizational challenges teams run into when AI-powered features become part of critical systems?

Let’s explore where teams struggle most, including latency, cost control, reliability, security, observability, and governance, and how these concerns shape decisions around orchestration, agents, workflows, and data access. Using AWS-based examples, you’ll learn about practical design patterns, real trade-offs, and common failure modes that are often overlooked in early prototypes.

You will leave with a clearer understanding of how to design generative AI systems that are not only impressive in demos, but resilient, maintainable, and safe to run in production.

Target audience: backend engineers, platform engineers, and software architects responsible for designing or operating production systems that incorporate generative AI.

Session format: architecture-focused talk with practical examples drawn from real-world production systems. The session emphasizes design decisions, trade-offs, and failure modes rather than demos or service walkthroughs.

Duration: available as a 45-minute or 30-minute session.

Level: intermediate to advanced. Familiarity with distributed systems and cloud-based application architectures is assumed. No prior hands-on experience with generative AI is required.