- AI coding tools like GitHub Copilot have transformed software development and productivity.
- AI assistants close experience gaps, although they can lead to less secure and error-prone code.
- Software engineering skills, such as creating intelligent architecture for code, will still be relevant.
Before graduating from university, Jacob Jackson founded the AI coding assistant TabNine in 2018. Jackson, a computer science student, wanted to reduce the repetitive, sometimes boring tasks a programmer might encounter.
“We were just trying to keep people from typing keys. There was no talk of designing or writing complete algorithms from artificial intelligence,” he said.
His startup eventually raised about $60 million and was acquired by Israeli company Codota in 2019. Jackson continued to work with AI, continuing to intern and work at OpenAI, the company that built ChatGPT.
In the last couple of years, there has been a boom in AI coding assistants. OpenAI’s competitors have launched advanced AI generation tools, such as GitHub Copilot and Anthropic’s Claude. The increased use of ChatGPT and AI coding assistants has changed the way software developers do their jobs. Using AI to write code can also close the experience gap between more young entry-level developers and senior developers, as it usually takes several years of work and even personal projects to get up to speed.
“There’s no such thing as new developers anymore, because AI basically makes everyone beyond that,” said Nicolas Gauvreau, who has worked as a developer in Canada for more than 20 years.
More than 97% of 2,000 respondents across the US, Brazil, Germany and India said they have used AI coding tools at work, according to a GitHub survey published in August. AI code suggestion generation tools can also increase the productivity of software developers by 26%, according to a study that analyzed data from Microsoft, Accenture and an anonymous Fortune 100 electronics manufacturing company.
Developers say the adoption of AI coding assistants will accelerate the field of software engineering rather than eliminate jobs, like how calculators sped up math calculations despite initial protests from teachers against its adoption.
While these tools can increase productivity, they can also present security issues that create more work for developers. According to a 2022 study led by Stanford University cryptography professor Dan Boneh, people using an AI assistant type significantly less secure code than those who do not have access to those tools. While AI assistants can speed up the coding process, they can create more errors that require human supervisor intervention.
AI assistants can empower coders
Most AI coding assistants focus on auto-completion, meaning the tool suggests code as the programmer types. Other language learning models (LLM) require rapid engineering, where the user can then apply AI suggestions code as a starting point for their idea, depending on the complexity of the problem they want to solve.
Before the advent of coding assistants, DeepAI founder Kevin Baragona always had a Google search engine window open in case he needed help solving a problem. Programmers often did research from sources like Stack Overflow, an online community forum where coders would share their solutions. Stack Overflow’s traffic has dropped since the rise of coding assistants.
“Every few minutes when you’re programming, it was like cheat code back then, and it just normalized like what you do when you’re coding: you Google a lot,” Baragona said.
Knowing more than one coding language, such as JavaScript, Python or Ruby, gives a programmer more flexibility in the job market when companies change their priorities. However, learning a completely new language would take a lot of time and learning.
Now, deep learning models have allowed more programming functions to be translated from one language to another, making it easier for developers to switch between programming languages without having to learn them quickly. Baragona said these tools make him feel like he knows “every programming language, even though I don’t, because IT will help me get over the hump very quickly.”
Gauvreau said AI coding assistants have empowered him because he feels less intimidated about taking on more clients even when he may not yet know the solution. He said he has doubled the number of languages he has learned in the last year, more than his entire career.
AI can help computer science students
Instead of shying away from coding assistants, some universities developed their own versions that would guide students to the right questions—another way AI tools can close the skills gap.
David Malan, a professor who oversees the popular CS50 Introduction to Computer Science course at Harvard University and online at edX, helped create the cs50.ai chatbot for the course. Malan said AI programs are “very willing to answer any and all of your questions, but not in a way that’s probably consistent with what a good teacher or teacher would prefer you to do.”
“The goal is to really teach students how to think and how to solve problems with the tools we currently have and will eventually have when it comes to real-world and software application,” Malan told Business Insider.
AI coding assistants can especially help online students in the classroom, who may not necessarily have the luxury of a teaching assistant, to have a “virtual tutor by their side,” Malan said.
HE has shortcomings
While Baragona said AI coding will become an everyday reality for the next generation of coders, he thinks it’s training programmers to be lazier, which could create problems they won’t know how to solve .
“You quickly get to a point where AI did all the work, but there are still bugs, and you don’t understand the code at all because you didn’t write it,” he said.
Once the code reaches a certain level of complexity, it discovers that the AI has dug a hole so deep that it cannot climb out.
“And at that point, you’re actually screwed because you can’t understand the code, you can’t fix it, and neither can the artificial intelligence,” he said.
Programmers for Microsoft studios have been encouraged to adopt Microsoft Copilot as a coding assistant, according to an Activision Blizzard contractor whose identity BI has confirmed and who asked to remain anonymous because he is not authorized to speak to the press. However, he said he has to be very specific when working with Copilot.
“AI doesn’t have a vision of what you’re trying to build because coding is really like building a building. AI can build you a small part,” he told BI. “We actively tried to use Copilot in our trials, but it just didn’t work.”
While many developers have chosen to delegate specific coding tasks to AI to reduce their workload, they say building a strong foundation in computer science and software engineering will continue to be important.
Software engineers don’t just code; they also solve problems and design systems. In this case, people still have an advantage.
“Artificial intelligence tools today, they don’t create thoughtful architectures like a human would. They code with short-term thinking,” Baragona said.