The race to deploy artificial intelligence (AI) engineering agents that can rival — and even replace — human software engineers is accelerating.
Last week, OpenAI CEO Sam Altman said that by the end of the year, the company’s most advanced AI reasoning model could rank first in competitive programming contests in the world.
During a recent Q&A panel at the University of Tokyo, he said the first version of its reasoning model ranked one millionth among competitive programmers globally. Then it reached 10,000th place. OpenAI’s latest and most powerful reasoning model — o3 — was able to reach 175th place.
“Maybe we’ll hit no. 1 by the end of the year,” Altman said. “That’s like an amazing rate of scale for more compute in this new paradigm, and we don’t see any signs of that stopping.”
Five days earlier, Meta CEO Mark Zuckerberg said 2025 would be the year “when it becomes possible to build an AI engineering agent that has coding and problem-solving abilities of around a good mid-level engineer.”
While Zuckerberg later clarified that this year would see groundwork for AI engineering agents with wide deployment coming in later years, he added that he believes “this is going to be a profound milestone and potentially one of the most important innovations in history.”
Cognition AI was ahead of the pack. In March, it introduced Devin, an AI software engineer agent capable of planning and executing complex engineering tasks, recalling context, learning over time and correcting itself.
A team at Answer.AI tested Devin and found that “the results were sobering. Out of 20 tasks, we had 14 failures, 3 successes … and 3 inconclusive results. Tasks that seemed similar to our early successes would fail in unexpected ways.”
While AI will continue to get better, experts tell PYMNTS that there’s more to the success of a software engineer than coding skills.
“Simply coding is not the primary value of a software engineer,” Jim Olsen, CTO of ModelOp, told PYMNTS. “Actually figuring out what the desired solution and outcome is, and the nuances of design for providing a flexible solution that will stand the test of time” are skills AI cannot replicate.
Software engineers working in AI require the combined expertise of software development, programming, data science and data engineering, Olsen said. And while AI can potentially replace low-level engineers, the problem now becomes who to train for mid- to senior-level engineering positions that will still be needed, he added.
Ilya Smirnov, head of the AI/ML department at Usetech, holds a similar view. “AI will not replace software developers anytime soon. Even with customization, specific use cases and wishful thinking, AI has too many limitations. Nevertheless, AI will change the way software engineers work.”
Smirnov noted that 70% of developers say they use AI-enabled development tools in their daily practice, which gives them an edge on tasks and increases productivity.
“AI for software development is already changing the way teams test, debug and document software. Developers are using AI to mediate communication with teammates, analysts, customers and clients,” he told PYMNTS. “In particular, AI can speed up the addition of new features, bug fixes and support requests.”
Smirnov said developers use AI to write large amounts of code and generate test coverage of a particular piece of code. As such, AI accelerates development and makes it more continuous, he said.
However, despite AI’s strengths, developers are still needed because of complex coding requirements.
Some projects impose often conflicting development requirements on developers that can’t be met, Smirnov said. Solving these challenges requires communication between the product owner, project manager and client. In these situations, AI cannot help because it lacks precedents in its training and circumstances change dynamically.
Moreover, AI cannot predict the development requirements of the organization, so the code created by the AI assistant may not meet the security and performance requirements of the company.
Smirnov noted that AI works best in a narrowly defined domain.
“It can create code that solves narrow, specific problems, but those solutions will not be well aligned with the entire product,” he said. “It needs a developer to stylistically and functionally harmonize all the software being developed.”
