GitHub Copilot Reveals AI's Future Role in Software Development

Software developers communicate using computer languages. Engineers are proficient in commands and symbols, relying on coding skills to build applications. Today, tools that support developers are evolving, making a new generation of engineers more like train dispatchers—relying on algorithms to turn natural language prompts into applications. With AI feedback, these tools promise to make building software applications fast and simple.

This is the core of Copilot, jointly created by GitHub and OpenAI. The platform is essentially an AI-enhanced assistant that acts as a virtual pair programmer for developers. Copilot is trained on public code repositories and is currently available as a technical preview.

GitHub Copilot uses OpenAI Codex—an engine based on GPT-3 that, according to its official website, "extracts context from comments and code and instantly suggests single lines of code and entire functions."

Although Copilot is not the only tool promising AI-assisted development workflows, Arun Batchu, senior research director at Gartner, noted that it could be a step forward because it leverages GPT-3's machine learning capabilities to translate natural language into code. To some extent, it can bridge the gap between business needs and the software being built.

"This is exactly what excites me about the potential of the new generation of tools," Batchu said. "They raise the bar."

As intelligent tools assist development, future software developers will focus more on connecting business needs with technical implementation. However, the tools heralding this shift have not yet reached the maturity needed for widespread use and trust. They must first ensure security before they can be widely adopted at the enterprise level.

AI, the dream pair programmer?

In the increasingly crowded market of developer tools, several solutions have previously offered features similar to GitHub. Providers such as Tabnine, Kite, and Codota leverage open-source codebases and libraries to generate suggestions in real time.

Usman Lakhani, research director at Info-Tech Research Group, said that other code completion tools allow rapid prototyping or creating "skeleton" applications, while Copilot aims to take over more of the "filling" process. "This is exactly where Copilot is trying to cut in," Lakhani said.

According to GitHub, although Copilot is trained on existing codebases, "the vast majority of its suggestions are uniquely generated and have never appeared before." The company declined interview requests but pointed questions to the product'sFAQ section。

Another market differentiator for Copilot is its adoption of GPT-3. This technology oncewrote an article describing itself, and its capabilities are orders of magnitude greater than its predecessor GPT-2. GPT-2 relied on 1.5 billion parameters to produce results, while the new model can leverage 175 billion parameters.

"From an AI perspective, especially the new generation of AI based on deep learning, just as it is enhancing or changing every other job, it is clearly also changing the job of software creation itself," Batchu said.

AI has hardly spared any traditional role. In the supply chain field, automation has reshapedhow warehouse workers operate. Robots have entered the marketing field, providing examples of effective marketing copy. AI also handles many customer service queries that were previously resolved by humans.

Tools like GitHub Copilot use AI to help engineers "discover the shape of the application you need to build, but it won't be perfect," Batchu said.

"Copilot guesses the exact code I want to write about one in ten times, and the rest of the time it suggests something either pretty good or completely off the mark," Francois Zaninotto, CEO and co-founder of Marmelab, commented in aHacker News post. "But when it guesses right, it feels like it's reading my mind."

Is the technology mature?

Copilot is part of a larger trend: AI as agenerative tool's potential, as proposed by Gartner. AI no longer stands on the sidelines but can autonomously produce, whether it's more efficient marketing copy, new delivery routes, new compounds with therapeutic potential, or, as Copilot does, application infrastructure and the code that drives it.

"I still think most companies, especially large ones, and possibly small ones too, overall, have not yet seen significant transformation from AI in the software development cycle," said Nitzan Mekel-Bobrov, Chief AI Officer at eBay.

Although versions of these applications have existed for a while, the industry is still evaluating AI's long-term impact on the software development process. "I think the initial application or first phase is actually applying AI at the edges of existing processes, rather than changing the entire process," Mekel-Bobrov said.

As for code output, there is a question: can algorithms apply engineering best practices when autonomously generating code? Lakhani accessed the product's technical preview and found that most of the code did not follow coding best practices. "I felt like I was initially looking for a needle in a haystack," he said, "but that doesn't mean it won't improve."

If AI pair programmers are widely adopted, security will be another factor executives will want to scrutinize. Unprotected code entering enterprise applications would cause trouble. One of the biggest risks of automatically generated code is that "they might satisfy nine out of ten good practices, but the one good practice they miss is exactly the one that exposes the company to the greatest risk," Lakhani said.

In one study, researchers gave Copilot 89 different scenarios, generating 1,692 programs. Upon review, about 40% of the codecontained cybersecurity vulnerabilities, reported researchers Hammond Pearce, Baleegh Ahmad, Benjamin Tan, Brendan Dolan-Gavitt, and Ramesh Karri.

But some argue that AI tools might actually make software development safer. Building software with a database of verified code can reduce the chance of inadvertently adding security vulnerabilities to applications through newly written internal code, said Dori Exterman, CTO of Incredibuild. "As a developer, I am to some extent more limited in understanding the security vulnerabilities that the code I generate might produce," Exterman said.

Beyond security concerns, the platform's flexibility and its efficient integration into existing coding workflows are also in question, as they are highly demanding on computational resources. "The problem with Copilot is that you still need advanced AI hardware because the model is so large and computationally intensive," Batchu said.

The future of programmers

What does it mean to be a software developer? The answer evolves with the tools of the industry. "So far, the trend is somewhat similar," Exterman said. "All software development ecosystems try to minimize the amount of coding work developers need to do. For example, starting with libraries 30 years ago, or even earlier."

Decades later, as AI-assisted tools mature, their integration into the software development process could have a lasting impact on the profession. Experts foresee a new type of software developer emerging, more assisted by AI. "I definitely think this is a shift toward generalists rather than specialists," Mekel-Bobrov said. "Language-specific requirements will relax or disappear entirely."

AI platforms like GitHub Copilot, once ready for enterprise deployment, will bridge the gap between human specifications and code execution. This will affect how companies view software engineers and the role they can play for the enterprise. "I think it will make software developers take on more of a software designer role, rather than focusing on implementation," Mekel-Bobrov said. "This will give them more time to spend on the creative aspects of software design, and less or no time on mechanical tasks."

With AI tools that can translate natural language into code snippets or even entire applications, companies can iterate faster. From the perspective of business departments relying on IT solutions, software developers will evolve into software problem solvers, Batchu said. "In many ways, software craftsmanship will begin to happen," Batchu said.