What you will learn in AI Agents in Action
Build, orchestrate, and deploy autonomous multi-agent systems
Chapter by chapter
Concepts explained in this book
- planningSequential planning is the organization of dependent actions into an ordered plan in which later actions follow earlier ones.
- AI agentAn agent system is a software system in which one or more AI agents perform tasks through reasoning, interaction, and access to tools or data.
- agent profileAn agent profile is a specification of an agent's identity, background, demographics, behavioral guidance, and task-oriented characteristics.
- prompt flowA visual editor is a graphical interface for inspecting and modifying a flow without directly editing its underlying configuration text.
- conversational aiA chatbot is a software system that conducts conversational interactions with users, typically without autonomous multi-step task execution.
- actionAn agent action is an operation an agent selects or performs to advance a task.
- continuous improvementA feedback loop is a control pattern in which outcome information is used to modify subsequent actions or plans.
- evaluationEvaluation is the assessment of a solution, action, or outcome against criteria or expected results.
- memoryAgent memory is a pattern for retaining and retrieving information arising from an agent's interactions with users, itself, and other systems.
- GPT AssistantGPT Assistants are OpenAI assistant systems that use GPT models and associated tools to perform tasks through interactions.
- semantic functionA semantic function is a function whose behavior is specified by a prompt or profile and executed through interaction with a large language model.
- personaA persona is an agent's behavioral and biographical characterization that guides how it responds and performs tasks.
- NexusNexus is the agent platform used in these exercises for creating, configuring, and testing agent profiles and actions.
- behavior treeA behavior tree is a hierarchical control pattern in which nodes represent conditions, actions, and control-flow choices for selecting and executing behavior.
- chain-of-thoughtReasoning is the process of deriving conclusions, plans, or solutions by relating information through intermediate steps.
- prompt engineeringPrompt engineering is the deliberate design and refinement of inputs to a language model to obtain desired and consistent outputs.
- multi-agent systemA multi-agent system is a system in which multiple specialized agents work and communicate together to solve problems.
- proxy agentAn agent-assistant proxy is an intermediary agent that mediates interaction between a user and another AI assistant or model.
- similarity searchKnowledge retrieval is the process of locating relevant information from a knowledge store for use in reasoning or generation.
- native functionA native function is a conventionally programmed function that can be invoked by an orchestration framework or language-model application.
- domain knowledgeKnowledge is information available to an agent for interpreting context, reasoning, and performing tasks.
- promptA system prompt is a high-priority instruction supplied to a language model to establish its behavior, role, and response constraints.
- toolA tool is an external or internal capability that an agent invokes to perform an operation or obtain information.
- AutoGenAutoGen is an open-source framework for building applications in which multiple AI agents collaborate.
- AutoGen StudioAutoGen Studio is a graphical development environment for designing and running multi-agent applications built with AutoGen.
- CrewAICrewAI is a framework for building multi-agent applications by structuring agents, roles, tasks, memory, and tools.
- pluginA plugin is a registered capability consisting of functions that a Semantic Kernel can expose for use in conversations or interactions.
- retrieval-augmented generationRetrieval augmented generation is a language-model pattern that retrieves relevant external information and incorporates it into generated responses.
- AgentOpsAgentOps is a platform for monitoring, tracing, and analyzing AI-agent executions, including their costs and behavior.
- cosine similarityCosine similarity is a measure of the angular similarity between two vectors, commonly used to compare the meanings represented by embeddings.
- code interpretation capabilityCode Interpreter is an assistant capability that executes code to analyze data and generate computational outputs such as plots.
- groundingGrounding is the assessment of how well a prompt or agent profile satisfies criteria for being constrained by relevant information or facts.
- rubricA rubric is a structured set of criteria used to specify or evaluate the quality of an agent profile or response.
- text chunkChunking is the technique of dividing a larger document or text collection into smaller units for processing or retrieval.
- agentic behavior treeAn agentic behavior tree is a behavior tree whose control nodes organize reasoning and actions performed by an AI agent or agent system.
- prompt versionA prompt variant is an alternative version of a prompt that can be tested against other versions for comparative results.
- language modelA large language model is a neural language model trained on extensive text data to generate and transform natural-language content.
- custom methodA custom action is an assistant-configured capability that invokes an external service through a defined interface.
- knowledge baseA knowledge base is an organized collection of information that a system can retrieve or use when answering questions and performing tasks.
- Semantic KernelSemantic Kernel is an open-source Microsoft software development kit for building AI applications that orchestrate language-model capabilities, plugins, and related services.
- back chainingBack chaining is a goal-directed reasoning technique that derives required subgoals, actions, and conditions by working backward from a desired outcome.
- GPT StoreThe GPT Store is an OpenAI catalog and discovery interface for searching and exploring GPT assistants.
- sequential processingSequential execution is the performance of actions or tasks one after another in a specified order.
- self-consistencySelf-consistency is a reasoning technique that samples multiple candidate reasoning paths and selects an answer supported by their agreement.
- chat completion modelA chat completion model is a language model designed to generate responses across iterative conversational exchanges.
- evaluation flowAn evaluation flow is a workflow that computes or aggregates assessments of prompts, outputs, or use cases.
- OpenAPI specificationAn OpenAPI specification is a machine-readable standard description of an HTTP API's endpoints, operations, parameters, and data structures.
- LM StudioLM Studio is a desktop application for running large language models locally and exposing them through a server interface.
- vectorizationVectorization is the transformation of data such as text into numerical vector representations.
- generationA generative model is a machine-learning model that creates new content from input data rather than only predicting labels or classifications.
- file uploadA file upload is an assistant capability for supplying documents or other files as knowledge to a configured language-model assistant.
- StreamlitStreamlit is an open-source Python framework for rapidly building interactive web applications for data science and machine learning.
- natural language interfaceA natural language interface is a software interface through which users or systems interact with functionality or data using ordinary human language.
- Tree-of-thoughtTree of thought prompting is a prompting technique that explores and evaluates multiple intermediate reasoning paths while constructing a solution.
- compressionMemory compression condenses stored memory information while retaining content useful for later retrieval.
- group chat collaborationA group chat is a conversational interaction pattern in which multiple participants exchange messages in a shared discussion.
- cosine distanceCosine distance is a vector distance measure defined as one minus cosine similarity, ranging from zero for most similar vectors to two for semantically opposite vectors.
- term frequency-inverse document frequencyTF-IDF is a term-weighting method that scores words by their frequency in a document relative to their frequency across a document collection.
- fine-tuningA fine-tuned model is a pretrained model further trained on additional data for a narrower behavior or domain.
- condition nodeA condition node is a behavior-tree leaf node that evaluates a predicate about the current state and reports its result.
- semantic memorySemantic memory is a memory form organized around meanings, including conceptual information and associated feelings or emotions.
- observabilityObservability is the collection and analysis of execution signals that reveal an agent system's behavior, performance, and failures.
- chain-of-thought promptingChain of thought prompting is a prompting technique that guides a language model to produce or internally follow intermediate reasoning steps before reaching an answer.
- vector databaseA vector database is a database optimized for storing numerical vector representations and retrieving records by vector-based similarity or distance.
- few-shot learningA few-shot model is a model that can learn or identify a target category from a small number of examples.
- hierarchical coordinationHierarchical processing is the organization and execution of agents or tasks through multiple levels of authority or coordination.
- embeddingA vector representation encodes an object or piece of information as a numerical vector.
- agent engineAn agent engine is a software component that powers an agent by connecting its base behavior to tools and language models.
- sequential plannerA sequential planner is a planner that organizes and executes actions in an ordered sequence to achieve a goal.
- sequence nodeA sequence node is a behavior-tree control node that processes its child nodes in order and succeeds only when the required children succeed.
- PlaygroundThe Playground is the browser-based interface used in this book to create, configure, and manage assistants.
- tool useFunction calling is an LLM interface capability that produces a structured request to invoke a named function with extracted arguments.
- context windowA context window is the maximum amount of tokenized input and related context that a language model can consider in one interaction.
- agent_action decoratorThe agent_action decorator is a code annotation that marks a function as an action to be discovered and loaded by the Playground.
- py_treespy_trees is a Python library for constructing, visualizing, and executing behavior trees.
- pipelinePrompt chaining is a technique that decomposes a problem into sequential prompts whose outputs are passed to later prompts for stepwise processing.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.
