Talk Overview:
You’ve likely experienced the inconsistent magic of AI prompts. This session moves beyond basic trial-and-error to a strategic, production-ready approach for DevOps and SRE teams. Discover how clear structure, deep context, and model-specific patterns transform your AI interactions from “kind of works” to “consistently delivers” — grounded in real incident response and postmortem workflows.
Key Topics Covered:
The anatomy of production prompts (system messages, context, constraints, delimiters) Core techniques: clarity, Chain-of-Thought, format constraints, and compression The AI-as-Coach method for iteratively refining your own prompts Advanced patterns: Tree of Thought, Self-Consistency, and ReAct for agentic investigations Applying prompt engineering to blameless postmortems and root cause analysis Model-specific guidance for GPT-5.4, Claude Opus 4.7, and Gemini 3.1 Pro
What You’ll Master:
Why vague prompts fail and the structural patterns that succeed Chain-of-Thought reasoning that prevents fabricated timelines and guessed root causes Format constraints and JSON schemas that make AI outputs system-integration ready Prompt compression techniques that cut tokens 70%+ without losing quality Using AI itself as a coach to refine prompts through rapid feedback loops ReAct-style investigation patterns for agentic SRE tooling
Who Should Attend:
DevOps engineers, SREs, platform engineers, and incident commanders who want AI that reliably accelerates their reliability workflows instead of fabricating convincing-sounding nonsense. Level: Intermediate - assumes basic familiarity with LLMs and incident response practices
