Join our daily and weekly newsletter for the latest updates and exclusive content on industry-leading AI coverage. learn more
Agent AI is a huge hit in multiple sectors, including application development and coding.
Today, Github has finally joined the Agent AI Party with the launch of GitHub Copilot Agent Mode. The promise of Agent AI in development is to allow developers to build more code with just a simple prompt. The new agent mode allows Copilot to iterate through its own code and fix errors automatically. Github previews Project Padawan, a completely autonomous software engineering agent that can handle the entire development task independently.
The new Agent AI marks the latest step in the multi-year evolution of the AI-powered coding development space that Github has helped pioneers. Github, owned by Microsoft, first previewed Github Copilot in 2021. General availability occurred in 2022. In the AI world, ChatGpt is the common name, and most people have heard of the term “generated AI.”
Github is steadily repeating Copilot. Initially, the service relied on the Openai Codex Large Language Model (LLM). In October 2024, users acquired the ability to choose from a variety of LLMs, including Claude of Humanity, Google’s Gemini 1.5, and Openai’s GPT4O. In addition to launching agent mode, GitHub also adds support for Gemini 2.0 Flash and Openai’s O3-Mini. Microsoft emphasizes agent AI, bringing together one of the largest AI agent ecosystems in the market.
AI that supports “peer programming”
The new Github Copilot Agent Mode service has shaken up the development environment as a series of rivals led primarily by startups. Cursor, Replit, Bolt and Lovable are all chasing the growing market for development powered by Github, which has helped them create.
When Github Copilot first appeared, it was placed as a pair programming tool paired with developers. Github is currently leaning towards the term peer programming because it employs agent AI.
“The developer team will soon be joined by a team of intelligent, increasingly sophisticated AI agents who act as peer programmers for everyday tasks,” said Thomas Dohmke, CEO of Github. “With today’s launch of Github Copilot Agent mode, developers can generate, refactor and deploy code across files in their organization’s codebase with a single prompt command.”
Technical breakdown: How Github’s new agent architecture works
Since its first debut, Github Copilot has provided a range of core features. Among them is intelligent code completion, which is the ability to suggest code snippets to execute a particular function. Copilot also acts as an assistant, allowing developers to enter natural language queries to generate code or get answers about a particular codebase. The system is intelligent, but still requires a non-trivial amount of human interaction.
Agent mode is beyond that. According to Github, the platform allows Copilot to iterate through its own output and the results of that output. This can greatly improve results and code output.
A detailed breakdown of agent mode operations is as follows:
Understanding and planning tasks:
If given a prompt, the agent mode not only generates code, but also analyzes the complete task requirements. According to Github, the system “can infer additional tasks that are not specified, but are also required for major work requests.”
Iterative execution:
The agent iterates both its own output and the results of that output. Continue repeating until all subtasks are complete.
Self-healing ability:
Automatically recognizes output errors. You can fix issues identified without developer intervention. Analyze runtime errors and implement fixes. Suggest and execute the required terminal commands.
Project Padawan brings “power” to development
Agent mode is certainly more powerful than basic Github Copilot operations, but it’s still not a fully automated experience.
To reach that full experience, Github is previewing Project Padawan. In popular culture, “Padawan” is a reference to the Jedi apprentice in the Star Wars science fiction franchise.
Project Padawan is built on agent mode and extends with more automation. In a blog post, Dohmke pointed out that Padawan allows users to assign issues to Github Copilot, and that the Agent AI system can handle the entire task. The tasks include code development, repository configuration, and human assignments to check the final code.
“In a way, you’ll be like a passenger as a contributor to all Github repositories,” Dohmke says.
Compare GitHub agents with other agent AI coding options
Github is, in some respects, a slower participant in the Agent AI Coding Race.
Cursor AI and Volt AI debuted their first AI agent in 2023, and Replette released their agent in 2024. These tools have been around for over a year to gain iterations, follow-ups and develop brand loyalty.
I personally have been experimenting with replica agents for the past few months. This week, the company brought technology to its mobile app. I don’t think this is a big deal, but that’s true. The ability to use simple prompts without the need for a complete desktop setup to build the software is powerful. Replit’s agents also provide AI prompt tuning to help generate the best possible code. The replica system runs completely in the cloud, and users like me don’t need to download anything.
Bolt doesn’t have a mobile app, but it has a very good web interface that makes it easy for beginners to get started. Cursors are a little more bulky in terms of including downloads, but they are a powerful tool for professional developers.
So, how do GitHub Copilot Agent Mode compare? GitHub is the de facto standard for today’s internet code repositories. Over 150 million developers use GitHub, including over 90% of Fortune 100 companies. According to the company, over 77,000 organizations have adopted Github Copilot. It makes the technology very sticky. Organizations that already rely heavily on Github and Copilot will not easily leave technology.
Compared to Replit and Bolt, Github Copilot Agent Mode is not a web-based feature, and at least not today. That preview is currently only available on GitHub Copilot in VS Code. While it certainly creates a small barrier to absolute beginner entries, the reality is that VS code is undoubtedly the most popular and widely used integrated development environment (IDE).
Developers are noisy bunch. So there are so many different programming languages and frameworks (it appears to be a new JavaScript framework that is emerging every other month). The bottom line is about comfort and workflow. For existing Github Copilot and VS Code users, the new agent mode brings much-needed features to help you improve your productivity. For those who are not stuck in the Github Copilot world, Agent Mode is extremely useful for conversations about Agent AI-driven coding tools that use Github Copilot.
Github Copilot Agent Mode is currently available in preview and requires a VS Code Insider for early adopters. GitHub does not yet provide pricing details or dates for general availability.
