Senior Engineer Asks on Hacker News When AI Agents Will Take Over Software Architecture
A Hacker News post asks when the human leaves the software design loop, and whether reinforcement learning on architecture data could end the need for hands-on programming skills.
The post, titled "Ask HN: What is the current trajectory of a senior engineer?", asked what prevents Anthropic, OpenAI and similar laboratories from internally building, and possibly buying, datasets of good and bad software architecture and applying reinforcement learning to them. The author noted that large language models are already trained with reinforcement learning on problems that can be scored as right or wrong, citing code and mathematics as examples.
He contrasted the state of the technology in 2024, when he described LLMs as "a fancy autocomplete", with current talk of building "software factories" run by agents. He asked what would stop that development, and why software architecture would still be hard for agents within a year or two.
The author wrote that he increasingly sees software developers dismissing this view and insisting that humans still have a place guiding agents and designing architecture. In his own assessment, the business need for actual software programmer skills would be "90% out" in a few years, and product leads and similar roles would be favored because they bring more people skills. He ended the post with the line "it was a good run".
The submission had received 9 points and 3 comments at the time it was captured. The post did not name employers, projects or specific models, and set out no proposal beyond the question of what currently limits reinforcement learning applied to architectural judgment.
Hacker News "Ask HN" threads are user-submitted questions that other readers answer in the site's comment section, so the substance of any response rests with those comments rather than with the original post.