Agent loops, tool calling, planning, memory, multi-agent systems, MCP, evaluation and common failure modes.
36 cards · basic cards · AI-written, checked twice. Edit anything.
- What is an agentic loop?
Show answer
A cycle where an agent perceives the environment, plans actions, executes them, and observes the results to inform the next iteration.- What is the perception step in an agent loop?
Show answer
The step where an agent observes and gathers information about its current state and environment.- What is the planning step in an agent loop?
Show answer
The step where an agent decides which actions to take based on its current state, goals, and available knowledge.- What is the action step in an agent loop?
Show answer
The step where an agent executes planned actions, often by calling tools or making changes to its environment.- What role does feedback play in agent loops?
Show answer
Feedback provides observations of action outcomes, enabling the agent to continue the loop and refine future decisions.- When does an agent loop typically terminate?
Show answer
When the agent achieves its goal, determines no further progress is possible, reaches a step limit, or encounters an unrecoverable error.- What is tool calling in the context of LLMs?
Show answer
A mechanism where a language model requests to invoke external functions or services by returning structured function calls instead of only text.- Why would an AI model call a tool instead of answering directly?
Show answer
To access real-time data, perform calculations, access external systems, verify information, or complete tasks beyond the model's training knowledge.- What is a tool schema?
Show answer
A formal specification describing a tool's name, purpose, input parameters with types, and expected output format.- How does a language model integrate tool output back into its reasoning?
Show answer
The tool's output is returned to the model as part of its context, allowing the model to interpret the result and decide on the next step.- How does a model decide when to call a tool?
Show answer
Based on learned patterns in training, the model recognizes when a tool would help accomplish its task and outputs a tool call in the correct format.- What should happen when a tool call fails?
Show answer
The error should be communicated back to the model, allowing it to retry with different parameters, try an alternative tool, or handle the failure gracefully.- What is the difference between chain of thought and planning?
Show answer
Chain of thought is step-by-step reasoning about a problem, while planning involves creating a sequence of actions to achieve a goal in the world.- What is hierarchical planning?
Show answer
A planning approach where complex goals are broken into subgoals, which are further decomposed into smaller actionable steps, creating a goal hierarchy.- What is plan execution?
Show answer
The process of carrying out the steps of a plan in order, typically by calling tools or taking actions in sequence.
FlashKeepers