What are agentic workflows?
Agentic workflows are LLM applications in which the model works through a task in several passes, using reflection, tools, planning or other agents, instead of answering once. Andrew Ng popularised the name in 2024, and it reaches far beyond code: his own examples include writing text and answering questions. It sits on a different axis from vibe coding, which is about whether a human reads the result.
Origin Andrew Ng popularised the name in 2024: his letter of 20 March (opens in a new tab) speaks of “agent workflows”, and his letter of 27 March (opens in a new tab) is the first of his to use and define “agentic workflows”, though no source shows he coined it.
- In Polish
- Agentowe przepływy pracy
- Closest ladder level
- All levels, free guide
- Also searched as
- agentic workflow, agent workflows, AI agentic workflows
Then $19.99 a month or $99.99 a year. Cancel from your account page. 30-day refund on a first purchase.
Why it matters.
The name matters because it moves quality from the model into the loop around it. On the HumanEval coding benchmark, Andrew Ng wrote on 20 March 2024, GPT-3.5 "wrapped in an agent loop" reached up to 95.1%, against 67.0% for GPT-4 answering zero-shot; the figures are his team’s analysis of other teams’ results. Anthropic’s later split gives the working vocabulary: in a workflow, "predefined code paths" orchestrate the model and its tools, while in an agent the model directs its own process. A loop with tests or a critic in it checks results before a human sees them.
Agentic workflows vs vibe coding
They sit on different axes: an agentic workflow is how a system calls a model, and vibe coding is how a human treats the result.
| Question | Vibe coding | Agentic workflows |
|---|---|---|
| What it names | A way of working: accept what the model writes without reading the diff | A way of designing an LLM application: reflection, tool use, planning, several agents |
| Who writes the code | The model | The LLM, in a loop; the output need not be code |
| Who checks it | Nobody reads the diff; the human watches the behaviour | The loop itself: reflection, unit tests or a critic agent |
| What stops the loop | The app appears to work | A fixed number of passes, or a passing check |
| Where the name came from | Andrej Karpathy, post on X, 2 February 2025 | Popularised by Andrew Ng in The Batch, March 2024; no source shows he coined it |
| Where it breaks | Beyond throwaway work, nothing replaces the reader | Multi-agent output is hard to predict when agents interact freely (Ng, 17 April 2024) |
From workflow to agent to orchestration
Anthropic’s definitions separate a workflow from an agent; a multi-agent system, built on agents, is the third step.
| Step | Who decides what runs next | In Anthropic’s words |
|---|---|---|
| Workflow | Your code | “Workflows are systems where LLMs and tools are orchestrated through predefined code paths.” |
| Agent | The model | “Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.” |
| Multi-agent orchestration | A lead agent, or a script, coordinating several agents | “a lead agent coordinates the process while delegating to specialized subagents that operate in parallel” |
How each tool does it.
Cursor, Claude Code and Codex each name this their own way. Each note carries the date it was checked.
Claude Code
A dynamic workflow puts the plan in code: a JavaScript script that Claude writes for the task and a background runtime executes decides what runs next, not Claude turn by turn.
Subagents and agent teams sit at the other end: Claude, or a lead agent, decides turn by turn what to spawn next.
Dynamic workflows guide, Claude Code 2.1.283, 2026-09-26
Codex
A pipeline is a chain of codex exec --output-schema FILE -o out.json calls, each stage reading the previous stage’s file.
Orchestration patterns guide, Codex 0.157.1, 2026-09-26
Questions about agentic workflows.
How are agentic workflows different from vibe coding?
They sit on different axes, and no source contrasts them directly; Ng’s letters predate the term vibe coding. An agentic workflow is how a system calls a model: it iterates, uses tools and checks. Vibe coding is how a human treats the result: accepting it unread. Our reading: a workflow moves the checking into the loop, and vibe coding removes the human check after the loop.
What is the difference between an agentic workflow and an AI agent?
Anthropic’s "Building effective agents" (19 December 2024) calls both "agentic systems" and splits them: "Workflows are systems where LLMs and tools are orchestrated through predefined code paths", while agents "dynamically direct their own processes and tool usage". In a workflow your code decides the next step; in an agent the model does.
Are agentic workflows only about writing code?
No. Ng’s letter of 27 March 2024 says self-reflection improves output "on a variety of tasks including producing code, writing text, and answering questions". The four patterns he named, Reflection, Tool Use, Planning and Multi-agent collaboration, are not specific to software. Coding agents are one application of the idea, and this glossary covers that part.
Is GitHub Agentic Workflows the same thing?
No. GitHub Agentic Workflows is a product, in technical preview since 13 February 2026: "GitHub Agentic Workflows let you automate repository tasks using AI agents that run within GitHub Actions." The generic term is older, from Ng’s letters of March 2024. Keep the two apart when you search.
The vocabulary of AI-driven development
Every name below is defined against vibe coding: who writes the code, who checks it, and what stops the loop.
Read the long-form guide: agentic engineering vs vibe coding
Sources.
The primary sources outside this site that this page relies on.
- Agentic Design Patterns Part 1: Four AI agent strategies that improve GPT-4 and GPT-3.5 performance (opens in a new tab) Andrew Ng, The Batch
- Agentic Design Patterns Part 2: Reflection (opens in a new tab) Andrew Ng, The Batch
- Agentic Design Patterns Part 4: Planning (opens in a new tab) Andrew Ng, The Batch
- Agentic Design Patterns Part 5: Multi-Agent Collaboration (opens in a new tab) Andrew Ng, The Batch
- Building effective agents (opens in a new tab) Anthropic, Anthropic
- How we built our multi-agent research system (opens in a new tab) Jeremy Hadfield, Barry Zhang, Kenneth Lien, Florian Scholz, Jeremy Fox and Daniel Ford, Anthropic Engineering
- GitHub Agentic Workflows are now in technical preview (opens in a new tab) GitHub, GitHub Changelog
Keep reading.
The guides that go deeper, and the terms and comparisons next to this one.
In the docs
Related terms
Comparisons
Read the guides in the same words.
Open every guide with the 7-day free trial. Each term is defined once here, and the guides use it the same way.
Then $19.99 a month or $99.99 a year. Cancel from your account page. 30-day refund on a first purchase.