This article is part of VentureBeat’s special issue, “AI at Scale: From Vision to Prosperity.” Read more from the issue here.

Three years ago AI-powered code development was commonplace GitHub Copilot.

GitHub’s AI-powered developer tool has surprised developers with its ability to help complete code and even generate new code. Now, as early as 2025, a dozen or more generative AI coding tools and services will be available from vendors large and small. AI-powered coding tools today provide sophisticated code generation and completion features, and support multiple programming languages ​​and deployment patterns.

A new class of software development tools has the potential to completely change how applications are built and delivered — or so many vendors claim. Some observers worry that these new tools will spell the end for professional coders as we know them.

What is the truth? How do these tools actually make an impact today? Where do they fall and where is the market headed by 2025?

“This past year, AI tools have become increasingly essential for developer productivity,” said Mario Rodriguez, chief product officer at GitHubtold VentureBeat.

The promise of business efficiency in gen AI-powered code development

So what can next gen AI-powered code development tools do today?

Rodriguez said that tools like GitHub Copilot can already make 30-50% of the code in some workflows. Tools also help automate repetitive tasks and help with debugging and learning. They can even serve as a thought partner to help developers go from idea to application in minutes.

“We also see that AI tools not only help developers write code faster, but also write better quality code,” Rodriguez said. “In our latest controlled developer study we found that code written with Copilot is not only easier to read but also more usable – it’s 56% more likely to pass unit tests.”

While GitHub Copilot was an early pioneer in the space, other recent entrants have seen similar gains. One of the hottest sellers in the space is Replitwhich developed an AI-agent approach to accelerate software development. According to Amjad Masad, CEO of Replit, gen AI-powered coding tools can make coding anywhere between 10-40% faster for professional engineers.

“The biggest beneficiaries are the front-end engineers, where there is a lot of boilerplate and repetition of work,” Masad told VentureBeat. “On the other hand, I think it will have less impact on low-level software engineers where you have to be careful with memory management and security.”

What is more exciting for Masad is not the impact of gen AI coding on existing developers, but the impact it can have on others.

“The most exciting thing, at least from Replit’s perspective, is that it enables non-engineers to become junior engineers,” Masad said. “Suddenly, anyone could create software with code. It would change the world.”

Certainly gen AI-powered coding tools have the potential to democratize development and improve the efficiency of professional developers.

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

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

dig (formerly Codium AI) has built a series of AI agent-driven enterprise application development tools. Friedman says that with automated AI tools, anyone can now build basic websites faster and with more personalization than traditional website builders can.

“However, for the complex business software that powers Fortune 5000 companies, AI is not yet capable of complete end-to-end automation,” Friedman said. “It goes beyond specific tasks, such as question answering in complex code, line completion, test execution and code reviews.”

Friedman argues that the main challenge lies in the complexity of business software. In his view, pure large language model (LLM) capabilities by themselves cannot handle this complexity.

“Simply using AI to generate more lines of code can degrade code quality — which is already a significant problem in enterprise settings,” Friedman said. “So the reason we haven’t seen massive adoption yet is because there are so many advances in technology, engineering and machine learning that need to be achieved in order for AI solutions to fully understand the complex business software.”

Friedman says Qodo addresses that issue by focusing on understanding complex code, indexing it, categorizing it and understanding organizational best practices to create meaningful tests and code reviews. .

Another barrier to wider adoption and deployment is legacy code. Brandon Jung, VP of the ecosystem of the gen AI development vendor Tabninetold VentureBeat that he sees a lack of quality data holding back the wider adoption of AI coding tools.

“For businesses, many have large, old code bases and that code is not well understood,” Jung said. “Data is always critical to machine learning and that’s no different with gen AI for code.”

Towards fully agent AI-driven code development by 2025

No single LLM can handle everything required for modern business software development. That’s why leading marketers are embracing an agent-based approach to AI.

Qodo’s Friedman expects that by 2025 features that seemed revolutionary in 2022 — such as autocomplete and simple code chat functions — will be commoditized.

“The real evolution will be toward specialized agent workflows — not one universal agent, but multiple specialists each excelling at specific tasks,” Friedman said. “By 2025 we will see the majority of specialized agents developed and deployed to the end, if there are enough of them, we will see the next inflection point, where agents can work together to create complex software .”

It’s a direction that Rodriguez also sees on GitHub. He expects that throughout 2025, AI tools will continue to evolve to help developers throughout the software life cycle. That’s more than just writing code; it also builds, deploys, tests, maintains and even repairs software. Humans will not be replaced in this process, they will be augmented by AI that can do things faster and more efficiently.

“This can be accomplished using AI agents, where developers have agents that help them with specific tasks at each step of the development process – and critically, an iterative feedback loop that keeps the developer in control at all times,” Rodriguez said.

In a world where gen AI-powered coding will become increasingly mainstream in 2025 and beyond, there is at least one difference that will be key for businesses. In Rodriguez’s view, that’s the integration of the platform.

“To be truly successful at scale, AI tooling must be integrated seamlessly into existing workflows,” said Rodriguez.