Generative AI Engineering
Understand agents, structured tool calls, MCP connections, orchestration, memory, planning, handoffs, and human approval controls.
12 matching terms
Agents
AI agent
A system that uses a model to choose and execute actions toward a goal, often across multiple steps.
Use an agent when the path cannot be fully predetermined and tools must be selected dynamically.
An agent reads a ticket, searches documentation, drafts a fix, and requests approval.Tools
Tool calling
The model's structured request for an application to run an external capability.
Use it for search, database access, calculations, file operations, and business APIs.
Call get_order_status with the verified order ID.Tools
Function calling
A tool-calling pattern where available functions are described with names and typed arguments.
Use it when your application exposes a controlled set of callable operations.
Function: create_calendar_event(title, start_time, attendees).Tools
Tool schema
A machine-readable definition of a tool's purpose, inputs, and constraints.
Write precise schemas so the model can select the correct tool and form valid arguments.
Define amount as a positive number and currency as a three-letter code.Connections
Model Context Protocol (MCP)
An open protocol for connecting AI applications to external tools, data, and reusable prompts.
Use it to integrate compatible capabilities through a shared client-server contract.
Connect the AI host to a documentation MCP server and expose its search tool.Connections
MCP host
The AI application that coordinates one or more MCP clients and controls permissions and user experience.
Use the term when describing where user consent, model interaction, and server connections are managed.
The desktop assistant acts as the MCP host for local and remote servers.Connections
MCP server
A program that exposes tools, resources, or prompts to MCP clients.
Build one when a data source or service should be reusable across compatible AI hosts.
The CRM MCP server exposes customer search as a read-only tool.Workflow
Orchestration
Coordination of models, tools, agents, state, and control flow within a workflow.
Use it when multiple capabilities must run in sequence, in parallel, or conditionally.
Orchestrate retrieval, analysis, approval, and publishing as separate stages.Agents
Agent memory
Stored information that an agent can reuse across steps or sessions.
Use scoped memory for durable preferences, task state, or prior decisions.
Store only necessary data and define retention, deletion, and access rules.
Remember the approved tone guide for this project, not unrelated personal conversations.Agents
Planning
Selecting and ordering intermediate actions before or during execution.
Use it for multi-step goals with dependencies or alternative paths.
Plan: inspect inputs, retrieve evidence, produce a draft, then run validation.Workflow
Handoff
Transfer of a task and its relevant context from one agent or component to another.
Use it when specialized agents own distinct responsibilities.
The triage agent hands the billing case to the refund agent with evidence and constraints.Human control
Human-in-the-loop (HITL)
A control pattern where a person reviews, corrects, or authorizes part of an AI workflow.
Require it for high-impact, irreversible, financial, legal, or externally visible actions.
Pause before sending the refund and ask an authorized employee to approve it.