Agentic AI vs AI agents vs generative AI: the differences
Agentic AI vs AI agents vs generative AI explained with Anthropic's definitions: workflows vs agents, how each works, examples, and when to use which.
A product review has three slides in a row titled "Our generative AI strategy," "Our AI agent," and "Agentic AI roadmap." Someone asks what the difference is. Nobody answers with confidence, because the terms are used loosely everywhere.
They do mean different things. This guide separates them using Anthropic's definitions from Building effective agents and Effective context engineering for AI agents, checked on 16 November 2026.
The short answer
| Generative AI | Agentic AI (workflow) | AI agent | |
|---|---|---|---|
| What it does | Produces content from a prompt | Runs a model through steps defined in code | Decides its own steps and tool use |
| Who controls the sequence | The person prompting | The developer's code | The model |
| Uses tools | Not necessarily | Yes, at fixed points | Yes, as it chooses |
| Example | Drafting an email | Classify ticket, then route, then draft reply | Investigate a bug across files and fix it |
| Predictability | High | High | Lower |
Generative AI
Generative AI produces content: text, code, images, audio. You prompt, it responds. Chat assistants answering questions and autocomplete suggesting the next line are generative AI. No loop, no actions in the world.
Agentic AI: the umbrella
Anthropic notes that "agent" means different things to different people, from fully autonomous systems to prescriptive implementations following predefined steps. It groups all of these as agentic systems, then draws one architectural line:
- Workflows: LLMs and tools orchestrated through predefined code paths.
- Agents: LLMs that dynamically direct their own processes and tool use.
So "agentic AI" covers both. A support pipeline that always classifies, then retrieves, then drafts is agentic, but it's a workflow, not an agent.
AI agents
An agent decides its own next step. Anthropic's later, simpler definition: an LLM autonomously using tools in a loop. It reads a result, decides what to do, acts, and repeats until the task is done.
Coding agents are the clearest example. Claude Code, asked to fix a failing test, reads the test, searches for the code, edits it, runs the test again, and keeps going. Nobody wrote that sequence in advance.
AI agent vs LLM
The LLM is the model. The agent is the system around it: tools, a loop, context management, permissions. Swap the model and the agent changes; swap the system and the same model behaves differently. See what an AI coding agent harness is.
When to use which
Anthropic's advice is to find the simplest solution and add complexity only when needed, which may mean no agentic system at all. Agentic systems trade latency and cost for better task performance.
| Task | Use |
|---|---|
| Draft, summarize, translate | Generative AI |
| A fixed, repeatable multi-step process | Workflow |
| Open-ended tasks where the steps depend on what you find | Agent |
| Coding across a real codebase | An agent, with review |
Workflows give predictability for well-defined tasks; agents fit when flexibility and model-driven decisions matter.
What agents need that generative AI doesn't
Because agents act, they need more around them:
- Context: the right information, at the right time, without flooding the window. See context engineering vs prompt engineering.
- Permissions: limits on what they can do unsupervised. See Claude Code permissions.
- Memory: what carries across sessions. See how to give an AI agent memory.
- Shared decisions: what the team has already decided, so the agent doesn't re-decide it.
Where Lockstep fits
Lockstep supplies that last piece for coding agents. It captures team decisions from PRDs, Slack, and pull requests, routes each to its owner for approval, and briefs Claude Code sessions on the ones that apply to the work. Any MCP client can query them too.
See how it works.
Sources checked 16 November 2026
- https://www.anthropic.com/engineering/building-effective-agents
- https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents
Related reading
For product managers, see how to use AI agents as a product manager. For how much to review agent output, see vibe coding vs agentic coding. For two popular open source agents, see Claude Code vs OpenClaw vs Hermes Agent. For how agents talk to each other, see A2A protocol example.
Frequently asked questions
- What is the difference between agentic AI and AI agents?
- Agentic AI is the broad category of systems where a language model takes actions with tools. AI agents are one kind: systems where the model decides its own steps. Anthropic calls the whole category agentic systems and splits it into workflows, which follow predefined code paths, and agents, which direct their own process.
- What is the difference between generative AI and agentic AI?
- Generative AI produces content, such as text, code, or images, in response to a prompt. Agentic AI uses a model to take actions toward a goal: calling tools, reading results, and deciding what to do next, often over many steps.
- What is the difference between an AI agent and an LLM?
- An LLM is the model: text in, text out. An agent is a system built around an LLM that gives it tools and a loop, so it can act and use the results. Anthropic's simple definition of an agent is an LLM autonomously using tools in a loop.
- What is the difference between an AI agent and a workflow?
- In a workflow, code decides the sequence of steps and the LLM fills in parts. In an agent, the LLM decides the sequence itself. Anthropic recommends workflows for predictable, well-defined tasks and agents when flexibility and model-driven decisions are needed.
- Is a coding assistant like Claude Code an AI agent?
- Yes. Claude Code reads files, runs commands, edits code, and decides its next step from the results, which fits Anthropic's definition of an agent. Inline autocomplete, by contrast, is generative AI without the agent loop.
Keep reading