So here’s an interesting piece of research, especially as ‘AI Agent’ is guaranteed to become cliche of the year 2025. Researchers from Carnegie Mellon (amazing how well robber barons have retconned their legacy, but I digress…) set up a business (The Agent Company) using only software AI agents to test whether they could produce any usable software. Min Choi with the accessible tweet thread
here and you can download the paper
here. Bottom line is: they are capable but lack common sense. Which makes sense, actually.