One agent, one job, better A
summary
Claim: breaking tasks into narrowly defined agents produces more accurate, consistent, and cheaper results. Builds an Langflow app wired to LM Studio models, runs Pytest + LangSmith evals against a math dataset, comparing three agent setups: single-agent (answers correct but verbose), single-agent forced “no explanation” (fails correctness), and multi-agent pipeline. Multi-agent returns concise correct answers across the dataset and integrates into CI tracing, proving the modular-agent workflow actually works.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:08Agentic systems problem space
- 3:13Why multi-agent systems
- 6:45What are evals
- 10:17Demo with Langflow and LM Studio
- 12:22Kitchen sink agents limits
- 13:47Multi-agent task breakdown
- 15:52Testing with LangSmith and Piest
- 21:53Live test results
- 24:59Code walkthrough
- 29:12Resources and closing
- 30:16Q&A
related talks
02h 22mArchitecting and Testing Controllable Agents
40mHow To Tame Your Agents?
30mAgents Building Agents
26mBuilding Agentic Workflows: Patterns for Orchestrating Intelligent Systems.
23mUnit testing LLM Agents
49mAI Agents: Collaboration in Real-Life Applications
34mSingle Agent vs Multi-Agent AI | Which One Should You Use?
32mBuilding Multi-Agent Systems: From Concepts to Collaboration