I remember when people used to say similar things about using ASM, and then about the craft of writing things in C instead of managed languages like Java.
At the end of the day most people will only care about how the tool is solving the problem and how cheaply. A cheap, slow, dirty solution today tends to win over a expensive, quick, elegant one next year.
Now there are still some people writing ASM, and a lot of them as a hobby. Maybe in a few years writing code from scratch will be seen in the same way, something very few people have to do in restricted situations, or as a pastime.
Writing code by typing on a keyboard will be just a hobby?
Sure, and who is supposed to understand the code written by AI when we retire? Since writing code by typing on a keyboard will apparently cease to exist, who will write prompts for an AI and put the code together?
Person: Hey AI, build me a website that does a, b and c.
AI: Here you go.
Person: Looks like magic to me, what do I do with all this text?
AI: Push it to Git and deploy it to a web server.
Person: What is a web server? What is a Git?
AI: ... let me google that for you.
Yeah, I'm just not seeing it play down as in the conversation above.
> Sure, and who is supposed to understand the code written by AI when we retire?
Why someone would need to? Do the product/business people who order creating something understand how it is done and what is Git, a webserver etc.? It is based on trust and if you can show the AI system can consistently achieve at least humanlike quality and speed on almost any development task then there is no need to have a technical person in the loop.
> So there could never be a new provider or a new protocol because AI wouldn't be able to use them or create them
On what do you base this? Is there some upper bound to the potential in AI reasoning that bounds it skill to creating anything more complex? I think it is on the contrary - it is humans who are bound by our biological and evolutionary hard limits, the machine is not.
Business already doesn't know what security is, they will jump head first into everything which allows them to get rid of those weird developer dudes which they have to cater to and give a lot of money.
I personally would also assume that there might be a new programming language AI will invent. Something faster, more optimized for AI.
I already coded a few times with chatgpt (4). Devin doesn't has to be perfect but its clear (in my opinion) that this will become better and better faster than we think.
Presumably, the AI would have access to just do all the git and web server stuff for you.. The bigger problem I see would be if the AI just refuses to give you what you ask for.
> A cheap, slow, dirty solution today tends to win over a expensive, quick, elegant one next year.
I disagree with this platitude, one reason being the sheer scale of the hidden infrastructure we rely on. Just looking databases alone (Postgres, SQLite, Redis etc.) shows us that reliable and performant solutions dominate over others. Many other examples in other fields like operating system, protocol implementations, cryptography.
It might be that you disagree on the basis of what you see in day-to-day B2B and B2C business cases where just solving the problem gets you paid, but then your statements should reflect that too.
At the end of the day most people will only care about how the tool is solving the problem and how cheaply. A cheap, slow, dirty solution today tends to win over a expensive, quick, elegant one next year.
Now there are still some people writing ASM, and a lot of them as a hobby. Maybe in a few years writing code from scratch will be seen in the same way, something very few people have to do in restricted situations, or as a pastime.