This article is part of a VentureBeat special, “AI at Scale: From Vision to Viability.” Read more about this story here.

Three years ago the development of AI code was just right GitHub Copilot.

GitHub’s AI-powered tool surprised developers with its ability to help complete code and generate new code. Now, as early as 2025, a dozen or so AI writing tools are available from vendors large and small. AI-powered writing tools now offer advanced code generation and completion methods, and support multiple programming languages ​​and delivery methods.

A new class of software development tools has the potential to revolutionize the way software is built and delivered — or so many vendors claim. Some observers worry that these new tools could spell the end of coders as we know them.

What is the truth? How are devices doing these days? Where are they failing and where is the market headed in 2025?

“In the last year, AI tools have become very important in the development of software,” said Mario Rodriguez, the company’s CEO. GitHubhe told VentureBeat.

Business performance promise for next gen AI code development

So what can next gen AI coding tools do?

Rodriguez said that tools like GitHub Copilot can already generate 30-50% of the code in some workflows. These tools can also help automate tasks and help eliminate mistakes and learning. They can act as partners to help builders go from idea to project in minutes.

“We’re also seeing AI tools not only help developers write code faster, but also help them write better code,” Rodriguez said. “In our most recent developer survey we found that code written with Copilot is easier to read and more functional – with a 56% chance of passing unit tests.”

Although GitHub Copilot is an early pioneer in this area, other recent entrants are also looking at the same thing. One of the hottest sellers in the space is Answerwhich has developed an AI-agent approach to accelerate software development. According to Amjad Masad, CEO of Replit, AI-powered next gen AI tools can generate documents anywhere between 10-40% faster for professional users.

“The biggest beneficiaries are the front-end engineers, where there is a lot of boilerplate and duplication of work,” Masad told VentureBeat. “On the other hand, I think it doesn’t really affect the programmers on the ground where you have to be careful with memory management and security.”

What is most interesting about Masad is not the impact of gen AI deployment on existing developers, but the impact it will have on others.

“The really exciting thing, especially from Replit, is that it can turn non-engineers into micro-engineers,” Masad said. “Suddenly, anyone can program and code. This could change the world.”

Of course gen AI powered writing tools have the potential to democratize development and improve the skills of developers.

That said, it’s not a panacea and has limitations, for now.

“For simple, remote projects, AI has made incredible progress,” Itamar Friedman, cofounder and CEO of Qodo, told VentureBeat.

Dig up (formerly Codium AI) is developing AI-powered business development tools. Friedman said that by using only AI tools, anyone can create websites faster and with more customization than web developers can.

“However, for the complex business applications that support Fortune 5000 companies, AI has yet to create the ultimate machine,” Friedman said. “It excels at specific tasks, such as answering questions about complex code, completing lines, creating tests and code reviews.”

Friedman said the biggest problem is the complexity of business software. In his opinion, the LLM alone cannot solve these problems.

“Just using AI to generate more lines of code can make the code worse — which is a big problem for businesses,” Friedman said. “The reason we’re not seeing much traction right now is because there are still advances in technology, engineering and machine learning that need to be made for AI solutions to understand complex business applications.”

Friedman said that Qodo is tackling this challenge by focusing on understanding complex code, documenting it, categorizing it and understanding organizational best practices to create meaningful tests and code reviews.

Another obstacle to mass adoption is legacy code deployment. Brandon Jung, VP of Ecosystem at gen AI Development Vender Tabninetold VentureBeat that he sees a lack of good knowledge holding back the adoption of AI tools for writing.

“For businesses, a lot of them have a lot of old code and that code doesn’t make sense,” Jung said. “Intelligence has always been important in machine learning and it’s no different with gen AI in code.”

Reach full AI-driven development in 2025

There is no single LLM that can address all the requirements of modern enterprise software development. This is why leading vendors have embraced the AI ​​approach.

Qodo’s Friedman expects that in 2025 things that seem to have changed in 2022 – like easy and simple social networking – will be changed.

“The real revolution is going to be in the field of specialists — not all assistants, but many specialists who are good at a particular job,” Friedman said. “In 2025 we will see many of these experts being developed and deployed until finally, when this is enough, we will see the next place, where the partners can collaborate to create complex programs.”

It’s a trend that GitHub’s Rodriguez also sees. He expects that in 2025, AI tools will continue to evolve to support programmers throughout the software development process. It’s more than writing code; It is rebuilding, deploying, testing, maintenance and software development. Humans will not replace this process, they will be supplemented by AI that will make things faster and more efficient.

“This will be achieved through the use of AI assistants, where developers have assistants that help them with specific tasks through each stage of development – and in a difficult, iterative process that keeps the developer in constant control,” Rodriguez. he said.

In a world where next-gen AI-powered coding will increase in 2025 and beyond, there’s one differentiator that could be important to businesses. In Rodriguez’s opinion, that’s platform integration.

“To be more effective, AI tools must integrate seamlessly with existing workflows,” Rodriguez said.