Stream: Nürburgring
Time: 12:00 - 12:45
On Monday, the AI assistant was brilliant. By Friday, it had become almost impossible to work with. It hadn't forgotten anything. That was the problem. Every conversation, instructions from Db2 manual, DBA runbook, incident ticket, SQL snippet, meeting note, and troubleshooting log had become part of its memory. Faced with a simple question — a return code, a locking symptom, a catalog lookup, a recovery step — it retrieved everything, struggled to separate signal from noise, and confidently produced answers built on outdated, irrelevant, or even contradictory information: last year’s workaround for this week’s subsystem, a different environment’s naming standards, a procedure that no longer matched the current Db2 release. The lesson? Enterprise AI doesn't become smarter by remembering more — it becomes smarter by remembering better. This session explores Context Engineering, one of the fastest-emerging disciplines in enterprise AI, through a deceptively simple question: What should an AI remember, what should it retrieve only when needed, and what should it intentionally forget? Rather than focusing on models or prompt engineering, we'll explore how modern AI agents use different forms of memory — working, episodic, semantic, and procedural — and relate them to concepts Db2 and enterprise engineers already know: active troubleshooting notes, historical incidents, trusted catalog and platform facts, operational runbooks, and recovery procedures. Along the way, we'll examine why retrieval often outperforms information hoarding, why "perfect memory" is rarely desirable around systems of record like Db2 for z/OS, and why context has become more important than prompts when building AI assistants that are accurate, explainable, and production-ready. Through practical Db2-themed enterprise scenarios and demonstrations, we'll bring these concepts to life — showing how changing only an agent's context and memory strategy, not the underlying model, can completely transform the quality, accuracy, and trustworthiness of its answers.
There is currently no attachment for An AI Agent That Remembered Too Much.
I'm working as a Lead Automation Engineer at BMC Software. With a strong background in AI, DevOps and the Automation, I bring a wealth of experience to the table. My expertise lies in designing and implementing automated solutions to enhance efficiency, reduce manual intervention, and ensure the delivery of high-quality software. I thrive on leveraging cutting-edge technologies to streamline workflows and contribute to the success of software development teams. Let's connect and explore how we can drive innovation together! Honored with - IBM Champion 2026, OMP Ambassador 2026, Planet Mainframe Influencer 2026.