Key takeaways
- Agentic AI is software that pursues a goal by planning, using tools and checking its own results across many steps.
- Anthropic separates workflows, where code sets the steps, from agents, where the model chooses them.
- Anthropic advises starting with the simplest solution because agents cost more and errors can compound.
Agentic AI is AI software that works toward a goal on its own, deciding what to do next, using tools and checking its results, instead of answering one prompt and stopping. Anthropic, OpenAI and Google Cloud all describe it in similar terms in their own documentation, and the details they stress differ in useful ways.
The short version: a chatbot answers. An agent acts, looks at what happened, and acts again until the job is done or it gets stuck.
What agentic AI means
The word comes from "agent." Google Cloud defines AI agents as software systems that use AI to pursue goals and complete tasks on behalf of users, with reasoning, planning, memory and some autonomy. "Agentic AI" is the name people use for systems built this way, and for the approach itself. For the single-agent basics, see our explainer what is an AI agent.
Anthropic's engineering post Building effective agents, published on December 19, 2024, draws the line most clearly. It separates two kinds of "agentic" systems:
- Workflows are systems where LLMs and tools are orchestrated "through predefined code paths." A developer wrote the steps.
- Agents are systems where LLMs "dynamically direct their own processes and tool usage." The model decides the steps.
Most products people call agentic today sit somewhere between the two.
How an agent works
The three vendors list slightly different parts, but the loop is the same: look at the situation, decide, act with a tool, look at the result.
Google Cloud names six features: reasoning, acting, observing, planning, collaborating and self-refining. It also lists memory (short-term, long-term, episodic and shared between agents) and tools, meaning functions or outside resources the agent can call.
OpenAI's agents guide says agents can plan and complete tasks using tools, work with other agents, and keep context across steps. Its building blocks are tools (function calling, MCP servers and hosted capabilities), skills, state, an execution environment such as a sandbox, and orchestration, which covers loop control and handoffs between agents. The OpenAI Agents SDK runs that loop inside your own application.
The tool part is where a standard like MCP matters. It lets an agent reach outside systems the same way every time. We explain it in what is MCP, the Model Context Protocol.
Workflows and agents: five patterns
Anthropic lists five common workflow patterns, which are the building blocks of most agentic systems:
| Pattern | What it does |
|---|---|
| Prompt chaining | Splits a task into steps. Each model call works on the previous output, with optional checks in between. |
| Routing | Classifies an input and sends it to a specialized follow-up task. |
| Parallelization | Runs model calls at the same time and combines the results, either by splitting a task or by voting. |
| Orchestrator-workers | A central model breaks a task into subtasks, hands them to worker models and merges the results. |
| Evaluator-optimizer | One model writes a response and another critiques it, in a loop. |
An agent, in Anthropic's terms, goes further: the model picks its own path with no fixed sequence.
Agentic AI versus a chatbot or assistant
Google Cloud compares agents, assistants and bots on its page. In its framing, a bot follows predefined rules. An assistant responds to requests, and the user makes the decisions. An agent is more autonomous: it pursues a goal, handles multi-step tasks and can act proactively. Google's own table calls this a spectrum of autonomy, and real products often mix the three.
When agents are worth using
Anthropic's advice is to start with the simplest solution and add complexity only when it clearly improves results. Agents suit open-ended problems where the number of steps is hard to predict and a fixed path can't be hardcoded. They also need a model you can trust to make decisions, because they run for many turns.
The costs are real. Anthropic notes that agents bring higher cost and the chance of compounding errors, where an early mistake carries into every later step. It recommends extensive testing in a sandbox. A well-defined task usually does better as a plain workflow.
What we don't know
None of the three sources gives a single agreed definition of "agentic AI." Google Cloud and OpenAI define or describe agents rather than the broader term, and Anthropic uses "agentic systems" as an umbrella over workflows and agents. Expect the word to keep meaning slightly different things in different product pages. Anthropic's post also says the tooling has changed a lot since December 2024, so check current docs before building.
Frequently asked questions
What is agentic AI in simple terms?
It is AI that works toward a goal on its own instead of answering one prompt. It plans steps, uses tools, looks at the results and keeps going until the task is done.
What is the difference between agentic AI and an AI agent?
An AI agent is one system that pursues goals using AI. Agentic AI is the broader term for systems and approaches built from agents. Anthropic uses 'agentic systems' to cover both workflows and agents.
What is the difference between agentic AI and generative AI?
Generative AI produces content such as text or images from a prompt. Agentic AI uses a generative model as its reasoning engine and adds tools, memory and a loop so it can act toward a goal.
What are examples of agentic AI patterns?
Anthropic lists prompt chaining, routing, parallelization, orchestrator-workers and evaluator-optimizer as common patterns, plus fully autonomous agents that choose their own steps.
Sources
- Building effective agentsAnthropic
- Agents guideOpenAI
- What are AI agents? Definition, examples, and typesGoogle Cloud
