It was said in a training today that “with AI, everyone is a developer now.” Fortunately I wasn’t in this particular training or I might have ended up in an argument with the trainer. In case you didn’t know, that’s bad form. Anyway, I disagree with this statement enough that I had to write about it.
There are numerous stories of lawyers providing legal documents which were AI generated citing law that doesn’t exist. There are medical papers which have been published with references to papers which were never penned. If AI can’t make a lawyer a lawyer or a doctor a doctor, it is highly unlikely AI will make someone who is not a developer into a developer.
I have actually witnessed junior engineers using AI to vibe code solutions which were horribly bug ridden. In one particular case, the resulting software was such a massive security liability that it had to be removed completely from service. In other words, AI can’t make a software developer into a software developer.
One does not simply build software
There is at least one service that I know of for certain that builds software based on prompts from users. The software is probably vaguely novel to the user, and it might even work. Software, however, has a lifespan. Generative AI has no sense of software lifespan without being directly prompted for it. Even then, there is only so much that can be done to address the need for ongoing maintenance of said software.
Software developers are, at the core, problem solvers. They take business requirements and figure out how to automate them in a way that is useful to the company directly, or their customers. AI doesn’t magically teach people how to do that. If you give a non-developer with a coding agent a feature spec and ask them to create it, it is likely to come out a mess.
There is more to software than writing code. I have already mentioned this once recently, but Ford learned this the hard way. They decided that AI was going to be a suitable replacement for knowledge workers at the company, so they let about 300 people go. Quality sank and business knowledge that had been core to the inner workings of Ford vanished as people vacated their offices and cubicles.
Software development requires that people understand and hold multiple things in their heads at once: business domain knowledge, technical knowledge, user requirements, feature specifications, and a working understanding of the codebase that is already in place. All of this is juggled atop the normal workplace interactions that define how people share knowledge. AI is capable of juggling multiple different aspects of the work, but only when it is provided by someone who has digested the information well enough to explain it.
Once software is written, there is ongoing maintenance that must happen, refactoring that will be done, and architecture which must be carved from the code. So, yes, AI is capable of cutting 1000 line pull requests, but is it capable of actually building and maintaining software without a trained developer? Unlikely.