What is prompt engineering?
Prompt engineering is the craft of writing and refining instructions for a language model so that it reliably produces the output you need. No coiner is established: the phrase was in use around GPT-3 in 2020. Since mid-2025 Anthropic, LangChain and Cognition have called context engineering its next step or its superset. It concerns the quality of the input; whether anyone reviews the output is a separate question.
Origin No single person coined it: OpenAI’s CLIP paper (opens in a new tab) of 26 February 2021 already refers to “the ‘prompt engineering’ discussion around GPT-3”, and Andrej Karpathy’s post (opens in a new tab) of 24 January 2023, “The hottest new programming language is English”, made the idea popular.
Also on Wikipedia (opens in a new tab)
- In Polish
- Inżynieria promptów
- Closest ladder level
- All levels, free guide
- Also searched as
- prompting, prompt design, prompt crafting, prompt programming, prompt engineer
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.
For coding agents, a prompt works as a brief: what to change, where, why, and a done-when check the agent runs itself. This site’s Claude Code guide puts the fix for a vague request this way: it is not a longer prompt but one that says what “done” looks like, so that the evidence arrives with the diff. That is where prompting meets verification. The prompt is also only part of what the agent reads, because instruction files, tool output and history count too, which is why the field now talks about context engineering.
Prompt engineering vs vibe coding
They sit on different axes: prompt engineering is about the quality of the input, and vibe coding is about the absence of review of the output.
| Compared on | Vibe coding | Prompt engineering |
|---|---|---|
| What it names | A way of working: accepting AI-written code without reading it | A craft: writing and refining instructions for a model. Also a 2023 job title |
| Who writes the code | The model | The model, from the prompt; the term does not say who checks it |
| Who checks it | Nobody reads the diff; the human checks behaviour by eye | The human reading the response |
| What stops the loop | “it mostly works” (Karpathy) | Nothing of its own: the human decides the answer is good enough |
| Where the name came from | Andrej Karpathy, post on X, 2 Feb 2025 | No coiner; in use around GPT-3 in 2020, in OpenAI’s CLIP paper by 26 Feb 2021 |
| Where it breaks | Past “throwaway weekend projects”, the scope Karpathy gave it himself | When the prompt is the only lever: a prompt that names its own check stops being vibe coding |
Prompt, context and harness: three questions
Each layer answers a different question. The first two questions are Alexey Grigorev’s wording from July 2026; the harness line is our reading.
| Layer | The question it answers | Who named it | How sources relate it to the others |
|---|---|---|---|
| Prompt engineering | “what we say when we interact with the agent” | No coiner; in use around GPT-3 in 2020 | Anthropic: context engineering is “the natural progression” of it. LangChain: it is “a subset of context engineering” |
| Context engineering | “what the agent knows before it starts” | Popularised by Tobi Lütke, 19 Jun 2025, and Andrej Karpathy, 25 Jun 2025; not coined by either | Cognition: “the next level” of prompt engineering |
| Harness engineering | What constrains and checks the agent | Mitchell Hashimoto, 5 Feb 2026; an OpenAI post popularised it, 11 Feb 2026 | Böckeler nests harness inside context engineering; Wikipedia’s “Agent harness” article nests context inside the harness. Neither is settled |
How each tool does it.
Cursor, Claude Code and Codex each name this their own way. Each note carries the date it was checked.
Cursor
A reusable prompt template gives the agent five things up front: the task, a canonical file to copy, the boundaries, the checks to run and the evidence to report.
The verification block matters most, because test and type results then judge the output instead of a line-by-line diff read.
Prompt templates for Cursor’s agent, 2026-10-02
Claude Code
A request is a short brief: what to change, where, why, and a done-when check that Claude runs itself.
The done-when clause matters most, because it turns a request into work Claude can prove with a test or command.
Prompt engineering for Claude Code, 2026-09-26
Codex
Each task gets a short brief with four parts: goal, context, constraints and a “done when” list of commands that must pass.
Durable project rules live in AGENTS.md, not in the prompt.
Prompt Codex with task briefs, plan mode and AGENTS.md, 2026-09-26
Questions about prompt engineering.
How is prompt engineering different from vibe coding?
They sit on different axes. Prompt engineering is about the quality of the input; vibe coding, a term Andrej Karpathy introduced on 2 February 2025, is about the absence of review of the output. You can vibe code with a careful prompt and review every line a careless one produces. No source sets the two side by side, so this is our reading. A prompt that names its own check stops being vibe coding.
Is prompt engineering dead?
As a label it is fading. Gartner’s summary of 28 July 2025 reads “Context engineering is in, and prompt engineering is out”, and Wikipedia notes that the job title of prompt engineer “has since become less common”. Vendors describe a successor rather than a death: Anthropic calls context engineering “the natural progression of prompt engineering”. The skill remains, because every agent starts from instructions.
What is the difference between prompt engineering and context engineering?
A prompt is the instruction; context is everything the model sees when it answers. Anthropic calls context engineering “the natural progression of prompt engineering”, LangChain argues that prompt engineering is a subset of it, and Cognition calls it “the next level”. Context engineering decides what the agent sees; harness engineering decides what checks and constrains it. Some authors treat one as part of the other.
Who coined prompt engineering?
No coiner is established, and Wikipedia’s article names none. OpenAI’s CLIP paper of 26 February 2021 already refers to “the ‘prompt engineering’ discussion around GPT-3”, so the phrase was established by early 2021. Andrej Karpathy’s post of 24 January 2023, “The hottest new programming language is English”, made the idea popular.
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.
- Learning Transferable Visual Models From Natural Language Supervision (opens in a new tab) Alec Radford et al., arXiv
- Post on X: “The hottest new programming language is English” (opens in a new tab) Andrej Karpathy, X
- Don’t Build Multi-Agents (opens in a new tab) Walden Yan, Cognition
- The rise of “context engineering” (opens in a new tab) LangChain, LangChain Blog
- AI-Native Development: Specifications, Loop and Graph Engineering (opens in a new tab) Alexey Grigorev, Alexey On Data
- Lead the Shift to Context Engineering as Prompt Engineering Fades (opens in a new tab) Gartner
- Effective context engineering for AI agents (opens in a new tab) Anthropic Applied AI team, Anthropic
Keep reading.
The guides that go deeper, and the terms and comparisons next to this one.
In the docs
- The full A-Z glossarySubscription
- Vibe coding vs agentic engineering: 21 terms comparedFree
- The autonomy ladder: which level is your workflow at?Free
- Prompt engineering for Claude CodeSubscription
- Prompt Codex with task briefs, plan mode and AGENTS.mdSubscription
- Effective prompting techniques for coding agentsSubscription
- Prompt templates for Cursor’s agentSubscription
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.