What is AI pair programming?

The concept of pair programming emerged as a formal coding practice in the 1990s, when Jim Coplien described “Developing in Pairs” in Pattern Languages of Program Design, but was discussed in computer science research as early as 1991.¹

The rise of pair programming is due to the boost in code quality and developer productivity it affords. With one developer dedicated to catching bugs and logic flaws, the other can code freely, allowing them to maintain focus and flow state. Many organizations are beginning to realize the same benefits in AI pair programming as well, when the second developer is not a human engineer but an AI coding assistant.

Coding tools that go beyond the simple command line were used even in the early days of computing, but it wasn’t until the 1990s and the ubiquity of the PC that visually rich coding environments with menus and toolbars were possible. In the following decades, these tools became more sophisticated to manage the growing complexity of code and the size of codebases.

Before the generative AI revolution of the 2020s, these coding tools were limited to rules-based automations and analysis. They lacked the deep understanding of code that was later made possible by AI. Microsoft’s Visual Studio Code (VS Code) is an example of a popular coding environment that offered such useful tools during this period.

The modern era of AI coding assistance began in 2021 with the release of GitHub Copilot, one of the first popular tools capable of generating code from natural-language prompts or partially written programs. The field expanded rapidly in late 2022 with the launch of ChatGPT, which focused on a more conversational experience.

During the same period, Amazon CodeWhisperer brought AI-powered code generation to AWS developers. In 2023, Cursor introduced an AI-first code editor that embedded conversational assistance directly into the development environment. Anthropic followed with Claude Code in 2025, an agentic coding assistant. Coding agents can not only generate code, but can perform tasks autonomously in codebases. IBM Bob is an example of the next phase of these tools—it extends AI assistance beyond coding itself to the full scope of enterprise software engineering workflows.

Today’s AI coding assistants are powered by large language models (LLMs) which use machine learning to generate new code based on their training data. This training data is made up of massive amounts of real-world code, giving AI-assistants enough domain knowledge to serve as competent pair programming partners.

Today, a coder can simply ask the assistant to perform tasks in natural language, and the AI behaves as a copilot, performing entire workflows or even developing apps from a single prompt. Most coding assistance, however, is a more iterative process involving human and AI collaboration. AI pair programming has rapidly become, in just a few years, a new standard mode of software development.

Real-time collaboration with an AI assistant is a good example of how AI acts as an augmentation to human workers. Software development often involves repetitive or tedious tasks such as writing boilerplate code, creating data models, formatting APIs or performing automated testing. AI assistants can complete many of these tasks in seconds. Software engineers can then spend more time focusing on architecture and design, embedding business logic into the code.

Using an AI pair programmer can significantly shorten development cycles and enable organizations to deliver software more quickly. Here are some of the coding tasks that an AI coding assistant can perform, just to name a few:

• Code generation

• Code completion

• Code review

• Debugging assistance

• Code refactoring

• Unit test generation

• Code documentation

• Security scanning

• Root-cause analysis

• Pull request review

• API integration

• Mitigation of technical debt

• Developer onboarding

• Performance optimization

• CI/CD pipeline creation

Many organizations think that AI pair programming is simply a way to make coding happen faster, and while this is true, developer efficiency is only one of the benefits. Enterprises and large organizations should think of AI coding assistants not simply as code generators, but as platforms for accelerating software delivery and institutionalizing engineering best practices across the software development lifecycle.

The strategic value is about much more than the individual coder’s productivity. Enterprises often struggle with silos, inconsistent practices and manual workflows. Some teams do things one way while another does things another way. Integrating AI pair programming across the organization allows the AI to serve as a source of truth on everything from onboarding to documentation. It’s a layer that potentially touches every department and every worker, gradually nudging everyone toward a single unified approach and integrated strategy.

In this way, AI pair programming is a massive improvement on two developers simply working together. Now one programming partner serves as the organization’s collective intelligence, pushing their many partners toward a more standardized and optimized way of working with code.

Organizations should establish policies governing how AI tools are used and what authority they themselves have to take action across codebases. Security, compliance, intellectual property protection and data privacy considerations should be addressed early in the adoption process.

Code governance is the organization’s set of rules and oversight mechanisms for software development. Organizations are now implementing governance policies that specify when AI-generated code can be used, how it must be reviewed and what validation steps are required before deployment. With the right policies and mechanisms in place, this new layer of code assistance will continue to revolutionize how software gets made.