You see it everywhere you turn:
"I built this using just AI and im making 10K USD MRR"
"Just got funded 1 million dollars at 100 million USD val...
For further actions, you may consider blocking this person and/or reporting abuse
Yeah anything remotely complex will still need dedicated devs to properly guide it - I don't believe in a completely non-technical business guy or CEO or whatever in building anything sophisticated which actually WORKS (and performs, and is secure) ...
That's why I'm also not that concerned about the prospect for us 'devs' ...
But, if people (non-devs) are able to vibe-code a not too complex app using AI, without spending a fortune - I say power to them!
the token spend is getting crazier nowadays though.
They may have to spend that fortune
Honest take from someone paid to fix things: cleaning up AI-built apps niche is real, but half the time the right answer isn't repair. It's telling the founder their app shouldn't exist in that form and rebuilding the approach from scratch. Knowing when to stop patching is the actual skill, and AI makes that decision harder to make, not easier.
This is so real... sometimes you have to suggest a change that makes you rebuild everything from the ground up .
Founders atimes don't like this.
e.g "using redis as your main database"... So many funny cases like that
I don't think this is a long-term niche. The AI is getting better and better at building apps.
Totally felt this.
AI can get an MVP surprisingly far, but once the app has real users, real edge cases, and real money flowing through it, the gap between “it works” and “it’s reliable” becomes very obvious.
The part about juniors worries me too. If entry-level work disappears, the whole pipeline gets weaker later.
Good post. It frames the situation in a much more realistic way than the usual “developers are finished” takes.
Non technical people are more exposed to programming now even though they don't really understand it well.
A lot of which chase the illusion of making so much money from a site built with one prompt.
So this is bound to happen.
Programmers are still revelvant.
This is true. I keep telling other developers as well to keep learning the fundamentals and learn deep. AI is not going to code itself, if it does, it's going to take more than 100% of Earth's energy and resources to compute ambiguous tasks.
In this day and age, AI will become the reaper of weak developers and an angel for strong and consistent developers. The bubble is close to bursting based from the numbers. When that time comes, make sure you are that strong and consistent engineer.
instead of using "..." to denote my pauses, i think i could do better with em dashes.
how do you type em dashes with your keyboard?
Maybe ask ai ?
I've been in the job market for around 6 months and I've been coding for over 20 years. I can't describe just how frustrating it is to be asked "do you have experience with AI?" Well, I apparently have enough experience to know I should be asking you the same.
I feel like those who don't know better don't want to. They prefer the comfortable lie to the uncomfortable truth. There have to be some of those non-technical folks who really, actually just don't know better. It feels like the decision-makers don't know what they're even looking for. Something fancy or shiny sounding?
I'm more than willing to give this "all-in on AI coding" a shot. I'm always looking to expand my perspective and problem solving toolset. I've just spent too much time disillusioned to the AI craze to cheer with everyone else.
Really. Im sick of developers. How much time do you really think is going to take the AI with XXXXX agents doing and iterating and reviewing XXXXX amount of times until is code perfect? Do you really think that special and unique? Its sickening to listen to a guy that has spent the las 20 years in a cave without looking at the sun talking about taste.
the gap between prototyping with AI and shipping production code is still huge. AI handles the first 70% fast - it is the last 30% where you need someone who can actually debug what the model got wrong.