From Prompt Hacking to Architectural Determinism: Engineering Reliable GenAI Systems
summary
Models fail when updates, context drift, and window compression change outputs, so the claim is to stop relying on prompts and instead make the LLM a router while delegating execution to deterministic, auditable code. The proposed pipeline captures full reasoning traces, uses human validation, codify success into executable artifacts (scripts/APIs), stores them in a skill store, runs them in sandboxes, and reuses validated fixes — an enterprise incident workflow that cuts repeated reasoning, lowers token cost, and makes remediation reproducible.
AI-written summary · about
chapters (9)
- 0:00Reliable GenAI systems intro
- 4:17Inconsistent GenAI outputs
- 10:42Prompt engineering limits
- 15:40Prompt to architecture shift
- 22:15System for deterministic outcomes
- 31:15Executable code vs prompts
- 39:21Reusable solutions flywheel
- 43:35Architectural approach recap
- 47:11Q&A
related talks
19mOn Engineering AI Systems that Endure The Bitter Lesson
20mBuilding Deterministic Infrastructure for Non Deterministic AI Agents
01h 39mMastering LLMs: From Prompt Engineering to Agentic AI
48mStop Fighting Your AI Engineering Prompts That Actually Work
31mModern Architecture for AI-Augmented Development Teams
29mBuilding Trust in AI Agents
54mDeterministic GenAI Outputs With Provenance
36mComposable Architecture for AI Systems: APIs, MCP, Skills, Sub-Agents, and Wor…