/Glossary · AI
What is AI agents?
AI agents are systems that use a language model to plan and carry out multi-step tasks, calling tools, APIs and data sources on their own. Unlike a chatbot, they take actions to reach a goal.
How AI agents work
An agent receives a goal, decides on the next step, calls a tool, observes the result and repeats until it is done. Tools can be internal APIs, database queries, knowledge base searches or actions in systems like a CRM. The model picks which tool to use based on its description.
On AWS you can build agents with Amazon Bedrock Agents or with frameworks like Strands Agents, LangGraph or CrewAI, and deploy them on Amazon Bedrock AgentCore. AgentCore provides an isolated runtime, memory, an MCP-compatible tool gateway, identity and observability, which are the hard parts to build yourself.
When agents make sense
Agents fit when a task requires pulling information from several systems and making intermediate decisions: resolving a support ticket, reconciling data, preparing a report or running an operational process with validations. If the flow is always the same and predictable, a deterministic workflow with Step Functions is usually cheaper and more reliable.
A good rule is to start with an agent scoped to one specific process, with few tools, a clear goal and human oversight on any action with real impact. Once evaluations show it is accurate, you expand its autonomy gradually.
Best practices and common mistakes
Design tools with precise descriptions and validated inputs, apply least privilege with an IAM role per tool, and require human approval before irreversible actions. Log every reasoning step and every tool call so you can audit them, and evaluate the agent against real cases before every prompt or model change.
Common mistakes: giving the agent broad access to production systems, wiring up dozens of tools that confuse the model, and not capping iterations or cost per run. Another risk is prompt injection through documents or web pages the agent reads as data.
/FAQ
Questions about AI agents
Another question? Write to hello@nubent.com and an engineer will answer.
What is the difference between a chatbot and an AI agent?
A chatbot answers questions from a prompt and, at most, a knowledge base. An agent plans steps, uses tools and takes actions in other systems to complete a goal, with some degree of autonomy and memory across interactions.
What is Amazon Bedrock AgentCore?
It is a set of AWS services for deploying and running agents in production with any framework or model. It includes an isolated runtime, memory, a gateway that exposes APIs as tools, identity management, a browser, a code interpreter and built-in observability.
What is MCP in AI agents?
Model Context Protocol is an open standard, created by Anthropic, for connecting models to tools and data sources in a consistent way. You expose an API once and different agents can use it without a custom integration for each one.
Are AI agents safe for production?
They can be, with the right controls: least privilege, input and output validation, Guardrails, human approval for critical actions, cost limits and full traceability. Without those controls, an agent can take the wrong actions or be manipulated with malicious instructions.
How we help with AI agents
AI Agents with Bedrock
Agents that carry out real tasks with Amazon Bedrock and AgentCore: tool use, MCP, Knowledge Bases, multi-agent orchestration, identity and observability.
Learn moreAI-native on AWS
Products and platforms with generative AI at the core: Amazon Bedrock, agents, RAG, evaluation, guardrails, LLMOps and inference costs under control.
Learn moreGenerative AI on AWS
Generative AI on AWS with Amazon Bedrock: assistants, RAG over your documents and GenAI features in your product, with evaluations, guardrails and cost control.
Learn moreMCP servers
Custom MCP servers that connect Claude and other agents to your internal systems and AWS, with authentication, least privilege, audit logging and deployment in your account.
Learn moreNext step
What should we build next on AWS?
30 minutes with an engineer. Concrete next steps, whether or not you work with us.