Rant: To the delusional engineers coping about AI
As a backend dev myself who’s been doing this for a while, reading comments on this thread made me a bit shocked. Are we really this arrogant? “Deploying things to production” cannot be our only sense of pride.
Half the comments are just engineers coping, acting like pushing code to AWS, wiring infra, or setting up a pipeline is some dark magic that non-techies can never figure out. Honestly, I’ve personally seen competent product guys use AI to build and deploy things from scratch that have easily scaled to thousands of users. They don’t need us to hold their hands for basic apps anymore.
Engineers in our ecosystem need a massive reality check. Deploying to production CANNOT be your long-term moat. It’s literally becoming a mechanical step now. Stop taking pride in this bullshit.
You need to start thinking about what to build, what not to build, and how to build it better than AI. System design, handling edge cases, making the right tradeoffs, solving actual business bottlenecks - maybe that’s what will save our jobs.
If you genuinely think “non-engineers cannot deploy” is your biggest power, you are going to struggle in the next 2 years. The market does not care about our ego.
One interview, 1000+ job opportunities
Take a 10-min AI interview to qualify for numerous real jobs auto-matched to your profile 🔑
You're wrong. It should invoke a sense of pride.
AI works on the back of public knowledge, i.e. GitHub repos and documentation. And it does not see and evaluate the real time traffic generated by apps and services.
Unfortunately, the reality is people are bad at documentation. Almost all infra products and components, including AWS, do not really work the way their documentation says.
It is easy to write a monorepo, but integrating systems at scale is just not an LLMs job. If you want to try your hand at getting an LLM to do that, go ahead and burn your wallet. But, there is such a thing as incompatibility. LLMs are not built for statistical/ time series analyses, and therefore are bad at evaluating tradoffs. You would just be throwing money at the problem with worse results than traditional statistical Machine Learning.
My real experience in using LLMs to try to debug platform level issues is that it just gives 5-6 highly unrelated reasonings, and most of the times none of them are correct.
Fair point. Keeping prod alive absolutely deserves pride.
But that’s not the same as saying non tech folks can’t ship anymore, because they are. The brag should be to be able to keep million user systems which are complex up and running even if they break.
No gatekeeping of course. But, look at the post you screenshot. They are from Walmart, Angel One, Flipkart. So, they are certainly working on high scale and pre-LLM systems. They are not deploying fresh apps. Your criticism doesn't hold.
I think you took this out of context. This is not arrogance but a reality. Don't take the literal meaning of deploying to production, that anyone can do. It means something different which devs know.
Here are my observations about AI:
- AI works fantastic when you have 4-5 files and mostly adding only new feature
- Say for UI, you have a React codebase with 100+ existing components, it doesn't work and try to add unnecessary code
- When the issue comes in code, it takes a lot more time to debug AI generated code compared to human written code
- it doesn't work properly to fix simple issues sometimes when context size increases
- it doesn't care about security, communication between micro-frontend based apps, different micro-services, reusability etc.
- it creates an illusion in the mind of non- engineers that software engineering is all about giving few prompts and everything should be available today.
- It increased workload on devs because not many people understand it's capabilities.
Maybe in 6 months it can do everything, i might change my observations. But today what non-techies.think vs what it actually does is lot different.
Though your points would make sense a couple of months ago , they don't make much sense now.
Currently working with a codebase with more than 40-50 files.
The context problem that you mentioned is almost already solved now with :
- CLAUDE.md
- Subagents / agent teams
They say garbage in - garbage out. So it's how you prompt it that matters here.
And regarding re-usability , best practices and what not we have something called plugins (you can also create skills files in your root dir) now which come built in with all these.
"Deploying to prod" is a phrase which implies stability. It's not about pushing the button and the actual upload.
AI is a great tool, but you are the one high in copium here.
The competent guys use AI as a tool, not as a blind magician creating things. The competent guy verifies the output and sends it for review under his name, not directly push to prod for millions of users.
this isn't just a vibe, the hiring patterns for backend roles are already shifting. the fear you're seeing is from people realizing their job description is being rewritten and they haven't read the new draft yet. your moat was never the tools, it was always the taste.
Aah yes, please speak when spoken to clanker
Why should "deploying things to production" without breaking things not hold a sense of pride? Non tech folks (even if we consider AI will build something magical which works initially) may scale the product to 100 or even 1000 users. But what beyond that? How will you scale while maintaining the codebase? One incorrect line of code which you deploy without verifying can break your whole system. Let's see you debug this with your non tech folks and get your system back up before even your 1000 users disappear and never use your app again.
Of course reliability, debugging, and scaling matter. But “I can deploy to prod” can not be your biggest argument against AI being able to build great software. I’ve seen AI-built products doing 10K+ daily active users without engineers hand-holding every step, and trust me, I know Swiggy is great scale - but most engineers in BLR aren't building stuff that crosses this scale today.
Deploying by itself is no longer a flex (which is what I had a problem with, from that thread)- but absolutely I agree, building for millions of users is something else altogether
I am fed up with these AI posts. There will be fewer jobs in software engineering, that's it. We accept it. And the folks who say software engineers are not needed, let LLM give you a wrong response and see its limitations. I still believe these LLMs are not the AI that can reason and produce something of its own.
Looks like someone is being insecure 😂
Blud thinks he’s cooking something
Are you a software engineer?
doesn't sound like one
Ah can't wait for this to unfold...