Agents vs Workflows: Why Not Both?
summary
Argues the agent vs workflow debate is overstated and shares concrete patterns: teams build pure workflow graphs for repeatable pipelines, others use ReAct-style agents for long-lived loops, and a third group combines them into nested workflows and handoffs. Recommends readable fluent APIs over graph node and edge APIs, shows breaking a single LM call into many improves reliability, demonstrates the agent supervisor model (agents calling agents) and dynamic tool injection to limit toolset size, and reports these hybrids actually reduce failures in production.
AI-written summary · about
chapters (9)
- 0:00Introduction: Agents vs. Workflows
- 1:00The Debate and Controversy
- 2:15"Don't Be That Guy": Critiquing Dogma in AI Development
- 3:40Harmful APIs and the Case for Fluent Syntax
- 8:00Defining Agents and Workflows
- 10:00Composition, Design Patterns, and Trade-offs
- 11:49Composing Agents and Workflows
- 12:12Architectural Patterns for Composition
- 14:43Q&A and Concluding Thoughts
related talks
35mAgentic Workflows: Stories from the Engine Room
14mOrchestrating Agentic Workflows with Governance and Trust
57mWhen AI Agent Teams Work (And When They Just Create More Work)
26mBuilding Agentic Workflows: Patterns for Orchestrating Intelligent Systems.
40mBuilding Effective AI Agents Simple, Composable Patterns
45mTaming AI Agents With Output.Ai
33mHumans on the loop, not in it: Taking agentic engineering end to end
50mHarnessing Event-Driven and Multi-Agent Architectures for Complex Workflows in Gen...