GenAI for Application Developers | Part 14 | Instrumentation and Tracing AI App | Langsmith

00:00 - The #1 Missing Skill in AI Engineering: Observability
01:50 - Traditional vs. AI Backends: Deterministic vs. Probabilistic
05:05 - The Cost Paradigm Shift: Fixed vs. Variable Token Costs
07:43 - Visualizing the Complexity: API Calls vs. Agent Loops
10:14 - The 4 Primitives of AI Tracing (Tokens, Prompts, Tools, Context)
17:16 - System Design: Architecting an Async Tracing Pipeline
20:00 - Terminology Deep Dive: What is a "Span" vs. a "Trace"?
24:15 - Performance Rule: Why Tracing Must Be Asynchronous
25:00 - Debugging Scenario: "Why did the bot send the wrong order?"
30:08 - Business Case: Identifying Users Who Bleed Your API Budget
32:28 - Code Walkthrough: Manual Instrumentation & Tagging
36:15 - Calculating Token Costs in Real-Time (Python Example)
44:10 - The Managed Solution: Introduction to LangSmith
45:45 - LangSmith Architecture: Background Workers & Queues
48:46 - Enterprise Risks: Data Privacy (PHI) & Vendor Lock-in
50:04 - Self-Hosted Alternative: LangFuse & ClickHouse
51:10 - Conclusion & Homework: Build a Cost Dashboard

Видео GenAI for Application Developers | Part 14 | Instrumentation and Tracing AI App | Langsmith канала Code And Joy