The InfoQ Podcast Context Engineering with Adi Polak
Apr 6, 2026
Adi Polak, Director at Confluent and author focused on streaming systems and ML engineering. He discusses context engineering as a stateful alternative to prompt-only approaches. Topics include turning prompts into reusable skills, building agentic workflows with long-term memory, using event streams like Kafka and Flink to enrich model context, and practical tips for experimenting with AI-enabled developer tooling.
AI Snips
Chapters
Transcript
Episode notes
Core Prompt Engineering Patterns
- Prompt engineering is about instructing models using roles, few-shot examples, chain-of-thought and constrained specs to get desired outputs.
- Adi Polak emphasizes these patterns teach models by example and rely on domain expertise to make prompts effective.
Cloud Agent Repaired Old Git Commit
- Adi Polak fixed an old Git commit issue by asking Claude to surgically remove a problematic line and push the corrected code within minutes.
- The anecdote shows agent tools can complete low-frequency, high-friction developer tasks much faster than manual context-switching.
Save Verified Workflows As Searchable Skills
- Build a searchable skills repository and save verified agent workflows as skills rather than retyping prompts each time.
- Adi Polak recommends stacking and selectively loading skills to avoid overwhelming model context and reduce cost.