More and more tech leaders seem to be saying that AI will be able to do programming roles end to end in the very near future.
Google Chief Scientist Jeff Dean says that AI will be able to operate at the level of a junior developer 24X7 in a year. “How far do you believe are we from having an AI operating 24X7 at the level of a junior developer?” Dean was asked at an event. Dean thought for a moment. “Not that far,” he replied.
“Is that six weeks or six years,” the questioner probed. “Every year in AI feels like 7 dog years or something,” Dean laughed. “I will claim that’s probably possible in the next year-ish,” he replied.
Jeff Dean is a towering figure in modern computing, known for his pioneering work at Google on large-scale distributed systems and artificial intelligence. He co-founded the Google Brain project and has led Google AI since 2018, becoming Google’s chief scientist in 2023 after the merger of DeepMind and Google Brain. His contributions include the development of MapReduce, Bigtable, Spanner, and TensorFlow, which have been foundational to modern AI and cloud computing technologies.
Jeff Dean isn’t the only leader in tech who’s talking about fully-autonomous AI engineers. Meta CEO Mark Zuckerberg had gone even further a few months ago, saying that AI could be as good as a mid-level engineer by the end of 2025. AI is already writing plenty of code supervised by human engineers at top companies — Google CEO Sundar Pichai says that well over 30 percent of the company’s code is now being written by AI, while Microsoft has pegged the number at around 20-30 percent. Anthropic, meanwhile, has said that 80 percent of the code for Claude Code is being written by Claude Code itself.
And apart from having the same capabilities as a software engineer, what will set these AI agents apart is their autonomy. Jeff Dean says that these systems will be able to able to operate “24X7”, which means that unlike human engineers, who’d need to sleep and take breaks for vacations, these AI agents would be able to work all the time. These AI agents could also be duplicated indefinitely, and as AI progresses, their capabilities would also increase. Also, it does seem likely that running these systems could be much cheaper than having human engineers — Devin, for instance, costs just $500 a month. It remains to be seen how these AI systems impact the job markets for programmers, but all indications point to the fact that it could end up being disrupted in a big way in the coming years.