We are at this weird stage where AI can replace engineers in some simple apps but the cost of doing so is so high that people immediately go back to hiring devs.
This stage is not going to last very long. Just hold your horses a bit.
Discussion
Indeed,
I think it takes another solid 3+ years before LLM's can actually replace devs for simple apps (in full)
Yeah, I truly think programming is a Bounded problem, like Chess, Go, etc. If we take that evolution, we should be able to have AI beating the smartest developers in about that time.
What do you suspect devs will do once that happens? Start doing more architecture work rather than lower level coding?
They are going to sell an art form, like people sell mechanical watches today.
If we don't need the people, will we even need human readable programming languages anymore? Just let the AI invent their own, or manipulate raw machine code.
Correct. We won't need compilers anymore.
Crazy idea. We could possibly think about training LLMs on pure binary data instead of natural language.
I think they are afraid of AI running on binary. But we already have AIs trained purely in voice tokens. Not voice translated to text and then inserted in the the AI. Pure voice. Takes in voice and replies as voice. Video native AI should be coming soon.
yeah cause pretty soon I will be able to craft an entire personality for an LLM and attach my identity to it then make an entire living off of unleashing it on the world to influence elections and shit
AI can't beat human creativity.
They can only expand on human creativity.
The real question is: Is human creativity a bounded problem or not. Meaning, if AI knows all the creative desires of every single human on earth, can it go beyond that or does it plateau?
Not so sure. Human creativity is just recycling information in novel ways. Computers already do the same thing.
LLM is a perfect tool for senior developers actually. Instead of managing 3 juniors, he can just delegate the task to AI.
Not today. Today you will be spending 3x the amount you can pay for a junior dev to get similar results. But yes.
The main issue is that without junior devs, there won't be senior devs in the future.
in my case, Sonnet comes handy when impementing specific functions that I know exactly how they should behave. And write tests for it. But asking AI to write 100+ LOC is spaghetti. The coding style and a train of thought is not human enough.
You got it right 👍..
#AI gets you three interns .. but the question is do you want to build stuff or train the interns :-) Besides coding ( in enterprise scenarios) is only ten percent of the job .. 90 percent is figuring out what to code and getting bunch of illiterate business users onboard !
In consumer apps - devs can do lot more with AI .. but they wouldn't have money to pay for #AI agents .. the 20 dollars monthly sticker is 99 % subsidized price .. if you amorotize the real cost , l price will be around 20 dollars and hour .. lot more than you would pay for an ace coder in India ..
So ..
The problem comes when you have to modify something that the AI built. If nobody knows how it works, it’s basically useless, right?
The AI can change it faster than any coders will ever do. Even if they know the code. That already happens today with Amethyst, for instance. The AI is faster than me.