The Evolution of Design Patterns in the Age of AI
summary
Argues that LLMs and agent swarms remain probabilistic, causing hallucination, token bloat and higher energy/costs; uses RAG, fine‑tuning and embedding tricks but shows limits: agentic proof attempts hit 35% success rate proving lemmas, clinical multimodal anomaly detection sits well under reliable thresholds, and business impact grows (cites $300 billion estimated cost of bad code). Recommends strict encapsulation, SOLID-style patterns to reduce context window, lower token use, and keep humans auditing to prevent codebase deterioration.
AI-written summary · about
chapters (8)
- 0:00Introduction and context
- 6:48How AI models work
- 15:27Limitations of current AI systems
- 23:16Impact on human cognition and workflows
- 28:16Design patterns for AI-assisted coding
- 36:57Validation and hallucination challenges
- 40:17Human role in the age of AI
- 42:44Q&A and closing remarks
related talks
01h 01mSoftware Architecture for Gen AI
32mAI Engineering Anti-patterns
40mBuilding Effective AI Agents Simple, Composable Patterns
27mEmerging Patterns Agentic AI
29mBuilding Trust in AI Agents
55mStop AI Agent Hallucinations: 5 Techniques + Production Patterns
52mThe Agentic Shift- Are We Reasoning, Or Just Retrieving Smarter?
01h 02mArchitecture Patterns for AI-Powered Applications