> Assuming software development becomes a commodity and the job becomes something like a fast food job where practically any adult who wants it can do it, what is your next move?
The assumption there is doing a lot of heavy lifting. You've presented something that's more like a "treatment" for a movie or TV show, rather than anything approximating what's actually happening with LLMs right now, or that can be seen to be happening in the near future based on current trends.
You're right that "[what is not debated is that LLM has changed our industry." Certainly, having a tool that can perform the task of reading all the prior knowledge and documentation and producing simple bits of code that hew to this documentation is transformative. How much someone says LLMs are transforming their workflows depends on how good they were at doing the above task themselves. I've never been good at reading manuals, and reading other people's code - for someone with my brain, having an LLM that can quickly and accurately (enough) answer pointed questions about very carefully constrained problem descriptions is a real boon. It scales me tremendously - projects I had given up on because I couldn't read up and act on the docs for all the different starting pieces have suddenly become something I can imagine at least getting started on.
I still have to plan the project, figure out all the questions at each step to ask about each piece, understand which of the LLM's concerns and objections are valid/relevant enough for me to pursue them, and understand the errors I continue to run into along the way enough to be able to state them clearly back into this "reference finding" mechanism.
The question of "software development" isn't turning into a despairing "What's there left for me to do?", but rather a much more hopeful, "Given that I am past the starting hurdles on so many projects simultaneously, which of these projects should I try to bring to some really interesting point of completion?" Now the constraining limit is more on how well I can conceptualize the underlying problem and less so on how quickly I can recall specific syntaxes for configuring a half-a-dozen different pieces of software.
Simple case in point: I wanted to throw up a web app inside a Docker container in an Ubuntu VPS. I just can't remember all the intermediate steps quickly enough - creating a secure enough deployment user/group setup; making sure Ubuntu is up-to-date; writing the docker-compose.yml file; figuring out what diagnostic commands to run to assess the various errors I run into along the way; working out my DNS and proxy server setup; and so on - even though each one of these is, in the grand scheme of things, a very basic step. Someone who needed to work through this list of steps, who has been creating micro-services using similar workflow templates in their company for 2-3 years, would have had little gain from using an LLM over just doing what to them is muscle memory. For me, it was the difference between, I guess I’ll just have to play with this idea as a toy on my laptop vs. Wow, this is running in the public Internet, with auto deploys from Github, and scheduled jobs running daily web scraping tasks for me while I sleep. And I could get there in maybe an hour’s worth of querying the LLM, reading error messages, and carefully triaging which ones to fix. It’s easy to underestimate just how much domain knowledge is still necessary even in this situation.
To go back to your analogy - now that you have a robot in your restaurant that can be fairly easily tasked by you to, say, make the perfect dough for a simple enough recipe; re-arrange your tables to add 3 more tops; be instructed to call all the butchers in town asking if they have specific cuts of meat; and so on and so forth - what can you, a star restaurateur, chef, food entrepeneur, do more of? It doesn’t mean that running that restaurant suddenly became as easy as the simplest of its composing tasks. The robot is flipping the burgers perfectly now so you can maybe save money hiring someone for that specific task. But the mere act of perfectly flipping burgers still leaves you many, many steps away from satisfying even one evening’s worth of diners.