Adi Polak, director at Confluent, discusses the evolution from prompt engineering to context engineering when building AI systems. Key themes include: why prompt engineering techniques like role assignment are becoming less effective; the importance of domain expertise for effective prompting; saving successful workflows as reusable skills to scale team knowledge; managing short-term vs. long-term memory in agentic systems; and how event-driven architectures using Kafka and Flink support stateful, multi-agent workflows. The conversation also covers RAG as one tool among many, and how agentic systems can automate engineering tasks like backlog triage and pull request generation.
Table of contents
Sponsored by GuardsquareTranscriptIntroduction [ 00:19 ]Defining Prompt Engineering vs. Context Engineering [ 00:43 ]Prompt Engineering as a Moving Target [ 03:36 ]From Prompts to Reusable Tools & Skills [ 06:05 ]Human-Centric vs. Agentic Workflows [ 07:01 ]Context Switching, Distraction, and Agent Assistance [ 10:38 ]Capturing Workflows as Skills [ 12:13 ]Context Switching, Distraction, and Agent Assistance [ 13:29 ]Developer Experience → Agent Experience [ 14:48 ]Short‑Term vs. Long‑Term Memory in Agent Systems [ 17:44 ]RAG as One Tool Among Many fo Multi-Agent Workflows [ 20:30 ]Event‑Driven Architectures with Kafka & Flink [ 23:56 ]Automating Backlogs and Engineering Workflows [ 26:42 ]Looking Ahead: Creativity as the Core Skill [ 29:13 ]About the AuthorSort: