I think Krugman is missing something here, with both electrification and computers/IT many new devices and software and interfaces had to be created from scratch for people to be able to use them, and people had to learn how to use the interfaces. With AI the interface is already there, it is just text or clicks sent to and received from computers and the infrastructure for this already exists. AI is plug and play with computer interfaces already and it is just software, so it could change everything (everything that is already interfaced to computers) very fast. We're already seeing lots of people promoting new applications they wrote in just a few weeks or months. There are still problems with the LLM's like hallucination and accuracy (though people have some problems with these also), but if the LLM's get better fast, then the impact will be almost immediate. For "good enough" applications the apps are already happening. How do others view the ease or difficulty of interfacing LLM's to existing systems? Would you trust an LLM to spend money for you, to schedule events, to interface to your bank and investment accounts, to control your house, to represent you to the public? They obviously could quite easily, but what would you actually trust them with? Trust here seems like a greater limiting factor currently than the speed of deployment.