Is the future of development just using AI agents to orchestrate everything and just reviewing, debugging, giving specs, verifying, etc...?
Rest of the information:
I’m a full-stack "developer" trying to understand my career in the age of AI.
I didn’t study CS, so I have a non-traditional background. I've also never worked with other developers until recently. I came from graphic design/web dev work (Wordpress, CSS, some JS, some PHP), did HTML email development for a bit (awful black magic), scripting/automations to help a content migration for a year, and now I have a full-stack role at a small company.
I now work with everything. Backend, frontend, infrastructure, CI/CD, integrations. Mostly "Hey this is broken" - my job is to fix it. I understand some of the fundamentals well enough to be productive. I can read code, improve it (usually), and mostly explain what needs to happen. I've been able to migrate their system into Azure on my own, handle some larger projects... I refactored their their payment system to integrate with Stripe. I've been pretty productive, to the point where I've gotten a promotion / title change.
But, I’m laughably bad at writing code. I feel like a total fraud. I came up in my early career at probably the worst time.
If you asked me to implement something simple without Claude or Codex, I’d struggle.
I was just getting started when ChatGPT came out, and since then I've never really needed to "learn to write code". It's mostly prompting and tweaking slightly from there. Things have been moving so fast I never really sat down and learned to write code. I have the books, I've read them, I've done some of the exercises, but obviously when I have a full time job, I'm mostly focused on that and keeping my personal life in order.
Being able to write code by hand and also being able to use AI agents gives you an added advantage over someone who, as you say, struggles to code without help of AI. Because you can do what that person can do, but they can't do what you can do.
as for the future development case, we don't really know what will happen, even until today agents still hallucinate and struggle to remember the context in the long form
atleast for me, i don't trust ai / agents 100% to completely detached myself on development, especially on reviewing and verifying part, they can cause more damage and time if something goes wrong, which almost certainly will, given some time
AI models are getting better and better and making fewer and fewer mistakes. However, in my opinion, you only truly master a skill once you’ve done it yourself.
It doesn’t really do you much good to rely solely on AI coding if you don’t understand the code. Experienced developers who have been writing code themselves for years can debug code much faster and more easily in many situations. AI models can make mistakes and sometimes fail to find those errors or produce incorrect results. If you don’t really understand the code, you’re pretty much on your own.
That’s why I would personally always recommend that you write code yourself from time to time, so you can gain a solid understanding of it.