What is AI pair programming?

AI pair programming is the framing of a coding assistant as a second programmer you work beside: the human steers and reads each suggestion as it appears. GitHub Copilot popularised the phrase “your AI pair programmer” on 29 June 2021, after Codota used it in 2017. On this site’s autonomy ladder it is Level 2.

Origin GitHub Copilot popularised the phrase in its launch post of 29 June 2021 (opens in a new tab), “Introducing GitHub Copilot: your AI pair programmer”, though Codota’s homepage already read “Your AI Pair Programmer” (opens in a new tab) in a Wayback Machine capture of 15 June 2017.

Last updated
By
Published
In Polish
Programowanie w parze z AI
Closest ladder level
Level 2, free guide
Also searched as
AI pair programmer, pair programming with AI, pairing with AI

Then $19.99 a month or $99.99 a year. Cancel from your account page. 30-day refund on a first purchase.

On this page Why it mattersAI pair programming vs vibe codingFrom pair programmer to peer: how the framing changedHow each tool does itQuestionsSources

Why it matters.

On this site’s ladder, pairing means reading every suggestion as it appears, the habit Level 2 depends on. Birgitta Böckeler, writing from inside the practice, warned in 2023 that framing assistants as pair programmers “is a disservice to the practice”. Microsoft, announcing a Copilot coding agent on 19 May 2025, described it as levelling up from “a pair programmer to peer”. Our reading: keep the pairing habit while you add the checks that let a loop run unread.

AI pair programming vs vibe coding

The two sit at opposite ends of review: in pairing the human reads every suggestion as it appears, and in vibe coding the human reads none.

Vibe coding compared with AI pair programming: what each names, who writes and who checks the code, what stops the loop, where the name came from and where it breaks
CriterionVibe codingAI pair programming
What it names A practice: prompt, glance at the result, accept it An interaction framing: the AI as a second programmer beside you
Who writes the code The model At first the human with AI suggestions of “whole lines or entire functions”; later the AI with the human steering
Who checks it Nobody reads the diff; the human watches behaviour The human, every line: “still reviewing every line” (Willison on Level 2)
What stops the loop “It mostly works” The human, suggestion by suggestion (our reading)
Where the name came from Andrej Karpathy, X post, 2 Feb 2025 Popularised by GitHub Copilot, 29 Jun 2021; Codota used it by 15 Jun 2017
Where it breaks When the software is for other people: Willison calls it “grossly irresponsible” there (6 May 2026) As a picture of pairing: Böckeler calls the framing “a disservice to the practice” (10 Aug 2023)

No primary source sets the two against each other directly, so the contrast in this table is our reading of how much each reads. Source: Vibe coding vs agentic engineering guide, 2026-10-10.

From pair programmer to peer: how the framing changed

The phrase predates Copilot, drew criticism from inside the practice, and was later outgrown by its main owner.

Dated uses of AI pair programmer from 2017 to 2026: Codota, GitHub Copilot, Birgitta Böckeler, Microsoft and Dan Shapiro, with what each said
DateSourceWhat it says
15 Jun 2017 Codota homepage (Wayback Machine) Heading: “Your AI Pair Programmer”
29 Jun 2021 GitHub, Nat Friedman “a new AI pair programmer that helps you write better code”, drawing context from the code you are working on
10 Aug 2023 Birgitta Böckeler, martinfowler.com “The framing of coding assistants as pair programmers is a disservice to the practice”
19 May 2025 Microsoft Azure blog The new coding agent “levels up GitHub Copilot from a pair programmer to peer”
23 Jan 2026 Dan Shapiro, Level 2 “You are pairing with the AI like a colleague.”

Source: Vibe coding vs agentic engineering guide, 2026-10-10.

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

This site’s Cursor guide pairs on one feature: the agent drives almost every keystroke, Plan Mode settles the design, and Agent mode builds one test-first slice at a time.

The developer moves to the navigator’s seat full-time: sets the direction, defines “done” for each step and checks the evidence. The page’s rule is that no slice ends on the agent’s word; each ends on a command you or CI can re-run.

Pair programming with Cursor’s agent, 2026-09-26

Questions about AI pair programming.

How is AI pair programming different from vibe coding?

They sit at opposite ends of review. In AI pair programming the human reads every suggestion as it appears; Simon Willison summarises Level 2 of Dan Shapiro’s ladder as “pair programming with the model but still reviewing every line.” In vibe coding, Karpathy wrote, “I don’t read the diffs anymore.” No primary source sets the two against each other, so this contrast is our reading.

Did GitHub invent the AI pair programmer?

No, it popularised the phrase. GitHub Copilot’s launch post of 29 June 2021 is titled “Introducing GitHub Copilot: your AI pair programmer”. Codota’s homepage already read “Your AI Pair Programmer” in a Wayback Machine capture of 15 June 2017. By 19 May 2025 Microsoft described its newer coding agent as levelling Copilot up “from a pair programmer to peer”.

Can an AI replace a human pair?

Birgitta Böckeler of Thoughtworks wrote on martinfowler.com on 10 August 2023 that she gets “quite frustrated when people talk about coding assistants as a replacement for pair programming”, and that the framing “reinforces the widespread simplified understanding and misconception of what the benefits of pairing are.” Our reading: AI pair programming is one interaction style inside AI-assisted coding.

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.