Context, Quality, Cost: The Truth About AI Agents | Tomasz Ducin

Building an impressive AI agent demo is one thing. Designing an agentic application that works reliably in practice is a very different challenge. Agentic AI systems are still software applications. They require architecture, data, queues, clear boundaries, and proven engineering patterns—not only prompts and access to a language model. In this AI Poland presentation, Tomasz Ducin explains the fundamental patterns behind practical agentic AI architecture. The session explores: • Why agentic applications still need solid software architecture • Patterns for working with databases and queues • How to manage limited context windows • How to address inconsistent model quality • Why infrastructure costs matter • Practical approaches to building more reliable AI systems • The difference between an AI demonstration and a production-ready application Tomasz Ducin is an independent consultant, software architect, developer, and trainer with experience across frontend and backend development. He has worked in the software industry since 2007 and specialises in JavaScript, architecture, and modern AI applications. This presentation was recorded at AI Poland. Learn more about AI Poland: https://ai-poland.pl/ Learn more about Tomasz Ducin: https://ai-poland.pl/speaker/2025/tom... What is the biggest challenge you have encountered while building AI agents: context, reliability, or cost? Share your experience in the comments and subscribe for more practical AI Poland presentations. #AIAgents #SoftwareArchitecture #AIPoland