What is agentic coding?
Agentic coding is a mode of work in which a coding agent plans a change, writes the code, runs it and iterates on the result in a loop, instead of suggesting the next line. No single coiner is established; Anthropic made it a product category with Claude Code in February 2025. The term says what the tool does and nothing about how carefully you check its output.
Origin No single coiner is established: the earliest use found is in Anthropic’s Claude 3.5 Sonnet announcement (opens in a new tab) of 21 June 2024, and Anthropic made it a product category on 24 February 2025 by introducing Claude Code as “a command line tool for agentic coding” (opens in a new tab).
- In Polish
- Kodowanie agentowe
- Closest ladder level
- All levels, free guide
- Also searched as
- agentic programming, AI agentic programming
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.
“Agentic coding” tells you what the tool does and nothing about whether the result is safe to merge. Two teams can run the same agent, one reading every diff and the other accepting all of it, and both are doing agentic coding. So ask the questions the name skips: what stops the loop, and who or what checks the result. Simon Willison put the practical side in October 2025: “If your project has a robust, comprehensive and stable test suite agentic coding tools can fly with it.”
Agentic coding vs vibe coding
The two are not opposites: agentic coding says what does the work, and vibe coding says how much you look at the result.
| Compared on | Vibe coding | Agentic coding |
|---|---|---|
| What it names | A way of working: accepting AI-written code without reading it | A mode of work of the tools: an agent plans, writes, runs and tests code in a loop |
| Who writes the code | The model | An agent, in a write-run-test loop |
| Who checks it | Nobody reads the diff; the human checks behaviour by eye | The term is silent: the agent’s own test runs, then a human, or no one |
| What stops the loop | “it mostly works” (Karpathy) | The agent’s goal, or a passing test run |
| Where the name came from | Andrej Karpathy, post on X, 2 Feb 2025 | No single coiner; earliest use found: Anthropic, 21 Jun 2024 |
| Where it breaks | Past “throwaway weekend projects”, the scope Karpathy gave it himself | When nobody checks what the agent wrote: it is then vibe coding with an agent |
Agentic coding vs agentic engineering
Agentic coding names what the tool does; agentic engineering names the discipline around it. The split is our reading, and one paper draws it from the other side.
| Compared on | Agentic coding | Agentic engineering |
|---|---|---|
| What it names | What the tool does: an agent writes, runs and tests code in a loop | The discipline of directing such agents without lowering the quality bar |
| Question it answers | What does the agent do? | What stays true about the software while it does? |
| Who uses the name | Anthropic and Cursor for their tools; Sapkota et al. and Hassan et al. in papers | Karpathy, Willison and Osmani, for the practice |
| Who is accountable | The term does not say | The human owns quality; tests and evals do the checking where they can |
Questions about agentic coding.
How is agentic coding different from vibe coding?
They are not opposites. Agentic coding says what does the work: an agent plans, writes, runs and tests code in a loop. Vibe coding, a term Andrej Karpathy introduced on 2 February 2025, says how much you look: you accept the output without reading it. You can vibe code with an agentic tool by accepting everything it does, and you can read every diff it produces. This contrast is our reading.
Is agentic coding the same as agentic engineering?
No. Agentic coding names what the tool does: an agent writes, runs and tests code in a loop. Agentic engineering names the discipline around it: directing such agents without lowering the quality bar, a name Andrej Karpathy popularised on 4 February 2026. Hassan et al. draw the same line in a 2025 paper, “elevating the practice from agentic coding to true agentic software engineering”. One is a mode of work, the other is how a professional uses it.
Is agentic coding the same as AI-assisted coding?
No. AI-assisted coding is the umbrella for every way of using AI to help write software, from autocomplete to agents to vibe coding. Agentic coding is the part of it where an agent runs its own loop. Wikipedia files it the same way: “Agentic coding” redirects to “AI-assisted software development”, defined as “the use of large language models (LLMs) and AI agents to assist software developers in software development”.
Who coined agentic coding?
No single person is established. The earliest use the research found is Anthropic’s 21 June 2024 announcement of Claude 3.5 Sonnet, which cites “an internal agentic coding evaluation”. Anthropic made it a product category on 24 February 2025 with Claude Code, and Sapkota, Roumeliotis and Karkee defined it in a paper dated 26 May 2025.
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.
- Introducing Claude 3.5 Sonnet (opens in a new tab) Anthropic
- Claude 3.7 Sonnet and Claude Code (opens in a new tab) Anthropic
- Vibe Coding vs. Agentic Coding: Fundamentals and Practical Implications of Agentic AI (opens in a new tab) Ranjan Sapkota, Konstantinos I. Roumeliotis and Manoj Karkee, arXiv
- Agentic Software Engineering: Foundational Pillars and a Research Roadmap (opens in a new tab) Ahmed E. Hassan et al., arXiv
- Introducing Cursor 2.0 and Composer (opens in a new tab) Cursor, Cursor Blog
- Vibe engineering (opens in a new tab) Simon Willison
Keep reading.
The guides that go deeper, and the terms and comparisons next to this one.
In the docs
Related terms
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.