Definition
Agentic workflows are pipelines where one or more AI agents take actions, call tools, and pass intermediate state between steps to complete a larger task. They occupy the middle ground between rigid pipelines (every step fixed) and fully open-ended agents (every step decided on the fly).
Episodes covering this
Worth reading next
Papers we haven't done a deep dive on yet, but would recommend on this topic.
- AFlow: Automating Agentic Workflow Generation
- InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents
- AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
- MAST: A Study of Multi-Agent LLM System Failures
- τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains
- LASER: LLM Agent with State-Space Exploration for Web Navigation
- AgentBench: Evaluating LLMs as Agents
- Automated Design of Agentic Systems
- AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery
- AgentLab: Scalable and Extensible Framework for Reproducible LLM Agent Experiments
- LEGO-Prover: Neural Theorem Proving with Growing Libraries
- AlphaEvolve: A Gemini-based coding agent for designing advanced algorithms
- AIDE: AI-Driven Exploration in Machine Learning Research
- Sagas
- Voyager: An Open-Ended Embodied Agent with Large Language Models
- AppWorld: A Controllable World of Apps and People for Benchmarking Interactive Coding Agents
- AppAgent: Multimodal Agents as Smartphone Users
- Cognitive Architectures for Language Agents
- Chain of Thought Empowers Transparent Reasoning of Language Models
- Reflexion: Language Agents with Verbal Reinforcement Learning