OpenAI's A-SWE: The AI Agent Poised to Transform Software Engineering

Will AI replace software engineers? It depends on who you ask

Artificial intelligence (AI) will soon be performing the essential tasks of software engineers -- or so the experts say. Sarah Friar, chief financial officer for OpenAI, proclaimed AI-as-software-engineer's emerging role at a recent Goldman Sachs conference. OpenAI's pending AI agent, called A-SWE (Agentic Software Engineer), "is not just augmenting the current software engineers in your workforce, but instead is literally an agentic software engineer that can build an app for you. It can take a [pull request] that you can give to any other engineer and go build it." Also: Why OpenAI's new AI agent tools could change how you code Not only does A-SWE build the app, but "it does all the things that software engineers hate to do, it does its own quality assurance, its own bug testing and bug bashing, and documentation," Friar continued. "Things that you could never get software engineers to do. So suddenly you can force multiply your software engineering workforce." With tools such as A-SWE emerging, should software developers and engineers be worried about their career prospects? Industry observers' reactions to the A-SWE initiative span the spectrum, from guarded pessimism to pragmatism. Also: The best AI for coding in 2025 (and what not to use) Software professionals "should be terrified," Andy Thurai, technology strategist, former analyst with Constellation Research, and former strategist with IBM Watson, told ZDNET. "The good ones will survive. The bad ones will be gone." Generative AI (Gen AI) "no longer just assists software developers and engineers; it's redefining the very nature of software development," agreed Lori Schafer, CEO at Digital Wave. "In the next five years, IT organizations will see a dramatic shift from teams of developers writing code line-by-line to leaner, more strategic teams of architects orchestrating AI-generated programs." What this trend means isn't necessarily wholesale job replacements, but a major shift in the roles and priorities of software professionals. "With AI agents producing fewer syntax errors, cleaner structure, and faster iterations, software developers and engineers are becoming editors and reviewers, not authors of every line," Schafer said. The rise of agentic AI in software "probably won't threaten job security per se immediately, but if you don't know how to use AI agents, then you might be threatened," Thurai pointed out. "Think of this: one person does this entire app in under a day, and the other takes four weeks to do the same thing. Who will survive longer? This trend will also mean fewer devs and software engineers will get hired." Also: AI agents aren't just assistants: How they're changing the future of work today Others suggested AI agents will augment rather than replace software development skills. OpenAI's A-SWE "represents a significant advancement in software development, but asserting that they can fully replace software engineers is an overstatement," said Neil Sahota, CEO at ACSILabs and AI advisor to the United Nations. "While A-SWE can write code, it doesn't understand the 'why' behind it. AI can mimic logic. However, it doesn't grasp context, business nuance, or edge cases that real-world systems need. Generating [create, read, update, and delete] operations is great, but it's a different approach to architect scalable, secure solutions under tight constraints." In large-scale enterprises or high-stakes domains, such as security, finance, healthcare, and compliance, "we'll have human software engineers in the loop for a long time to come," said Cassie Kozyrkov, CEO of Kozyr and former chief decision scientist and data scientist at Google. Software engineering "requires more than just the raw ability to understand and write code," said John Callery-Coyne, co-founder and chief product and technology officer at ReflexAI. "When AI companies are running these model benchmarks, they're usually working in a vacuum, but real-life software engineering doesn't happen in a silo." Also: As AI agents multiply, IT becomes the new HR department Effective software development requires "deep collaboration with other stakeholders, including researchers, designers, and product managers, who are all giving input, often in real time," said Callery-Colyne. "Dialogues around nuanced product and user information will occur, and that context must be infused into creating better code, which is something AI simply cannot do." The area where AIs and agents have been successful so far, "is that they don't work with customers directly, but instead assist the most expensive part of any IT, the programmers and software engineers," Thurai pointed out. "While the accuracy has improved over the years, Gen AI is still not 100% accurate. But based on my conversations with many enterprise developers, the technology cuts down coding time tremendously. This is especially true for junior to mid-senior level developers." AI software agents may be most helpful "when developers are racing against time during a major incident, to roll out a fixed code quickly, and have the systems back up and running," Thurai added. "But if the code is deployed in production as is, then it adds to tech debt and could eventually make the situation worse over the years, many incidents later." Also: Will synthetic data derail generative AI's momentum or be the breakthrough we need? In addition, the new roles of software professionals in an age of AI and agents need to be explored. "Where performance matters, software engineering agents are unlikely to eliminate the work -- they'll just shift it from writing the code to explaining and reviewing it, which isn't always a win," Kozyrkov said. It's likely that software professionals "will find themselves playing archeologist in the AI's mistakes," Kozyrkov added. "Most coders will tell you it's far more fun and fulfilling to write code yourself than read someone else's. AI-generated labor at scale sounds great on paper, but someone will still need to monitor the bots, fix their mistakes, evaluate edge cases, maintain long-term systems, and ultimately take responsibility. Unless we're careful, we risk replacing builders with babysitters. It's up to us how that plays out."

OpenAI Really Might be Coming for Your Coding Job

Friar painted a broad picture of OpenAI's wide-ranging ambitions, of which A-SWE forms a big part, as the company starts moving beyond providing AI models to customers into a service for more comprehensive AI infrastructure. The A-SWE agent is the third major agent-based AI model from the company's, the news site explained. The first is its Deep Research system, designed to help scientists (or finance workers, legal staff, or project managers) kick off new projects by carrying out some of the more mechanistic parts of the job like summarizing existing documents published on a particular topic. The second is its Operator tool, an agent that can autonomously carry out online tasks, potentially freeing up office staff to work on revenue-earning jobs more efficiently. Last week the Norway-based browser company Opera demonstrated a tool that works like this live on stage, wowing the audience of journalists and influencers. There is a debate over whether or not AI's can actually "write" code, rather than regurgitating and repurposing material they've "learned" by observing millions of lines of existing computer code carefully crafted by human experts. But it's clear that AI is transforming the industry, regardless of how its capabilities are described. A recent report, for example, highlighted worries over the coding chops of new graduates, since it found that some freshly minted computer science degree holders were already so reliant on AI to help them solve coding problems that they didn't actually understand the fundamentals of the products they're delivering.

OpenAI Is Creating AI to Do 'All the Things That Software Engineers Hate to Do'

In short, A-SWE will be able to take over the duties of software engineers. Can AI create an app with just a prompt? ChatGPT creator OpenAI is working toward that future by developing an AI agent that can replace crucial work currently done by human software engineers. At a Goldman Sachs conference last month, OpenAI's Chief Financial Officer Sarah Friar said that OpenAI is preparing an AI agent that goes beyond helping software engineers -- it does all of their duties for them. Instead of simply completing lines of code or generating new code, the agent acts on its own to create apps, run quality assurance tests, test for bugs, and write documentation. "It's literally an agentic software engineer that can build an app for you," Friar said. "Not only does it build it, it does all the things that software engineers hate to do. So suddenly you can force multiply your software engineering workforce." The AI agent, called A-SWE, will be OpenAI's third agent tool. In January, the AI startup released Operator, an AI agent that can shop for groceries or make dinner reservations. One month later, the AI startup launched Deep Research, an AI agent that surfs the web to generate in-depth research reports. Related: 'Maybe We Do Need Less Software Engineers': Sam Altman Says Mastering AI Tools Is the New 'Learn to Code' OpenAI isn't the first to develop an AI software engineer. In March 2024, Cognition AI released a coding assistant named Devin, claiming that it was the first AI software engineer. However, the AI was only able to complete three out of 20 coding tasks in an independent test. As AI's coding skills improve, industry leaders are warning that it may take over software engineering roles. Anthropic CEO Dario Amodei stated last month that AI could write "essentially all of the code" within a year, expressing "a fair amount of concern" about the impact of AI on jobs. Meta CEO Mark Zuckerberg told Joe Rogan in January that Meta is working on an AI that writes code as well as "a midlevel engineer" this year. Meanwhile, Google indicated in October that AI wrote one-fourth of "all new code" at the company. Related: Amazon Cloud CEO Predicts a Future Where Most Software Engineers Don't Code -- and AI Does It Instead OpenAI CEO Sam Altman said last month that the skills needed to excel in tech have changed because of AI. He stated that "the most obvious tactical thing is just get really good at using AI tools" instead of mastering how to code. "At some point, yeah, maybe we do need less software engineers," Altman said.