More and more tech leaders seem to believe that software engineering jobs are set to be disrupted in the not too distant future.
Replit CEO Amjad Masad is the latest to make a bold prediction, stating that AI will be able to fully replicate the work of a software engineer by 2027. His confidence stems from the rapid advancements in Large Language Models (LLMs) and their growing ability to perform complex coding tasks.
“When we built the agent,” Masad explained, referencing Replit’s Ghostwriter coding assistant, “it was a huge bet on the future of LLMs. We feel that LLMs are going to get insanely good this year. It’s going to be mind-blowing.”
Masad points to a specific benchmark called SWE Bench to illustrate his point. “The x-axis is the years and the y-axis is how many tasks can AI do that resemble software engineering tasks. These tasks are pulled from GitHub open-source repositories, tasks that humans have done, and now we’re testing AI’s ability to do these very tough open-source tasks.”
He highlights recent progress: “Yesterday, Anthropic announced their latest model, and it became the state of the art on this benchmark. You can see it’s around 70%, I think. So you can see this climb from 2022 where we had GPT-3 all the way to 2025 where we’re at 70%.” This rapid progress fuels Masad’s conviction. “If you project that forward,” he continues, “even by the end of the year, we’re going to get really far. So when it gets to 2027, and it saturates the benchmark, we have evidence that it’s autonomous enough to replicate a software engineer’s job.”
This isn’t just theoretical for Masad. He sees tangible, near-term implications. “So what does that mean? We can build software from scratch using agents. I’ll make one prediction: within a year, agents will be able to build 20% of SaaS software.”
Masad’s prediction comes at a time of increasing anxiety surrounding AI’s potential impact on the job market. OpenAI CEO Sam Altman has said that an AI will be the world’s best programmer by 2025, and former Stability AI CEO Emad Mostaque has said that AI is already better than outsourced Indian programmers. While some see these tools as empowering assistants, others worry about the potential for widespread job displacement. Masad’s 2027 prediction throws this debate into sharper relief, suggesting a much faster timeline for potential disruption than many anticipate. Whether his prediction proves accurate remains to be seen, but the rapid pace of LLM development certainly warrants close attention.