Practice
Make it stick: look up the jargon, flip the cards, then test yourself. Everything lives in this session only.
- Agentic system
- Umbrella term for both workflows and agents — any system where an LLM helps drive a multi-step task.
- Workflow
- LLMs and tools orchestrated through predefined code paths you wrote.
- Agent
- An LLM that dynamically directs its own process and tool use to accomplish a task.
- Augmented LLM
- The shared building block: a model augmented with retrieval, tools, and memory.
- Gate
- A programmatic check between chained steps that validates the intermediate result.
- Sectioning
- A parallelization variant: split a task into independent subtasks run at once.
- Voting
- A parallelization variant: run the same task several times and take consensus.
- Orchestrator
- The central LLM that splits a task into subtasks at run time and synthesizes results.
- ReAct
- Reasoning + acting: a Thought → Action → Observation loop (Yao et al., 2023).
- Reflexion
- Verbal self-reflection that improves an agent across iterations (Shinn et al., 2023).
- Handoff
- In a decentralized multi-agent system, one agent transfers control to another.
- Manager pattern
- A central manager agent coordinates specialized agents via tool calls (agents as tools).
- Guardrails
- Explicit limits and checks that keep an agent operating safely within bounds.
- ACI
- Agent-computer interface: the tool docs/signatures an agent uses — design it carefully.