Agentic workflow: when AI works in several steps
An agentic workflow is a way of using an AI model over several steps: it plans, calls tools, drafts, reviews and corrects, instead of producing its answer in one go. It is how assistants that “get things done” for you actually work.
Data as of October 22, 2026
Key figures
| Level | Intermediate |
|---|---|
| Origin | 2024 |
| Introduced by | Andrew Ng (popularisation, lettre The Batch du 20 mars 2024) |
From first draft to working in loops
When you ask an assistant a question, the large language model usually writes its answer straight through, without going back. An agentic workflow changes that: the task is broken into steps, and the model can look something up, try, notice a mistake and try again. Andrew Ng popularized the phrase in his March 20, 2024 letter, describing four patterns: reflection (reviewing its own output), tool use, planning, and collaboration between several agents.
A concrete example
Picture a consultant preparing a briefing on a competitor.
- Plan: the model lists the questions to cover (product, public pricing, recent news).
- Search: through tool calling, it runs web searches and reads the pages it finds.
- Draft: it writes a structured first version.
- Check: a second pass compares each claim with the sources and flags anything unsupported.
- Fix: the draft is revised, then handed to the consultant.
Each step is simple on its own; chaining them is what makes the result more reliable than a single answer.
Why it matters
This is the shift from AI that answers to AI that carries out work. Anthropic’s guide “Building effective agents”, published in December 2024, lays out common patterns: prompt chaining, routing a request to the right handler, parallelizing, letting an “orchestrator” coordinate workers, and having a second call evaluate the result. These building blocks show up in most automation tools today. Before handing a sensitive task to such a system, our Can I trust an agent with this? tool helps you weigh the risks.
Common mistakes and myths
- “More steps is always better.” Every step adds cost, latency and another chance to go wrong. Anthropic’s guide recommends starting with the simplest solution that works.
- “Agentic means autonomous, so no supervision needed.” For irreversible actions (payments, sending emails, deleting files), a human sign-off is still the safe choice.
- “It always involves several agents.” A single model can follow an agentic workflow; multi-agent setups are just one pattern among others.
Related terms: multi-agent system, human-in-the-loop, AI agent, orchestration.
Related terms
- AI agent · An AI that does more than answer: it plans and carries out a series of actions with tools to reach a goal you set.
- Tool calling (function calling) · The ability of an AI model to request that an external function be run with specific arguments, then use the result in its answer.
- Multi-agent system · A setup where several AI agents, each with its own role, split a task between them and share their results to get it done.
- Human-in-the-loop · A design where a person reviews, corrects or stops the important steps of an AI system or agent before they take effect.
- Agent orchestration · Coordinating several model calls, tools or AI agents: who does what, in which order, and how the results are brought together.
- Task decomposition · Breaking a complex goal into smaller subtasks that a model or AI agent can handle one at a time or in parallel.
Try it
Full glossary
Frequently asked questions
What's the difference between an agentic workflow and an AI agent?
The terms overlap. In its December 2024 guide, Anthropic calls a “workflow” a system whose steps are fixed in advance by code, and an “agent” a system where the model decides what to do next. In everyday use, “agentic workflow” often covers both.
Does an agentic workflow always give better results?
Not always. Extra steps often improve quality on complex tasks, but they cost more tokens and more time. For a simple question, a single request is usually enough.
Do I need to code to build an agentic workflow?
No. No-code tools such as n8n or Make let you chain steps around an AI model. Code frameworks give you more control for advanced cases.
Where does the phrase come from?
Andrew Ng popularized it in his March 20, 2024 letter in The Batch, which contrasted one-shot prompting with iterative work. He described four patterns: reflection, tool use, planning and multi-agent collaboration.
Data sources
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.