| > I don’t think there are any genuinely new capabilities that AI agents would need in order to take my job. They’d just have to get better and more reliable at doing the things they can already do. The "just" here is minimizing what has been the crux of the problem for the past ~5 years. This technology has been capable of producing code all this time. The end result has been improving due to massive scaling efforts, and some relatively trivial engineering ("reasoning", "agents", etc.). And yet reliability is still a massive problem. The tools still hallucinate, still lead the user in dead-end directions, and still do so confidently and randomly, without any discernible reason. Expert users are able to guide them to a certain extent, but whether the prompting incantation is done manually or via the trendy Markdown file of the week, it's all guesswork based on feelings and anecdata. I'm personally not too worried about being replaced by these tools, even though my skillset is nothing remarkable. My opportunities might shrink, but this is a two-way street. Companies that use "AI" indiscriminately don't interest me either. The demand for quality human work and ingenuity will always exist, even within a sea of mediocrity. I'm much more concerned with the societal impact of the mountains of shoddy software being produced, deployed into increasingly more critical infrastructure, and put into hands of incompetent and malicious people. There is very little thought and discussion on this topic, let alone any guardrails. "AI" companies are now attracting governments and advertisers, both full of malicious and incompetent people. The next decade is going to be interesting, that's for sure. |