Building with AI ? be careful

The unknown impact

When you are at a job, you are supposed to ship things end to end , faster but with fewer errors, So using ai for coding feels like the way to. but there are concequences of using it aggressively

1. Coding in a job and coding interviews are different.

When you are at a job, you are marking tickets as completed on daily bases , meetings, standups etc. but when it comes to interviews the interviewer will take your juice of by asking the internals, the concepts , even the acronyms are asked.

When you are vibe-coding ( i hate this term btw ) you are offloading the thinking part, the coding part, the system design part to the AI.

It does build stuff, when you look into the codebase, you can see it creates reusable functions, but some compoennts are long like 4000 lines of code, some are short , some part is good some part is not.

When you offload these things to AI. you forget the syntax , the architecture, you will not be able to explain your own project to the interview board. This is bad. because you cannot be stuck in this one company for the next 3 decades. you have to move, you have to learn stuff by yourself.

This is where the offloading part you did , just so that you can avoid the mental pressure for a little bit hits back.

Now you have a solid project in your hand , but coding a new feature on it by hand ??? Can you think about it ??

Why does this schema looks like this ?? When actual people use this app , they will encounter 100s of issues how will you handle that ? did you even test the feature ?, if someone is super good , they can immediately bypass your auth system, your payments etc. in no time. so when you dont know about your code, the architecture how will you secure it ? how will you know the flaws until someone ruin your app in production ?

So do not use AI at all.

No, that's as bad as well

1. you will end up spending 3 days for a feature or bug that can be fixed in 1 hour with AI.
2. you will make progression slower, even if you are super good. you cannot match the speed of morons using ai. the manager will notice it, because they know how better AI is at coding.

So whats now ?

How to use AI for coding

This is a future proof , battle tested learning path for you. even if something else comes in like AGI. you can still be super chill about it.

In 2023 when chatgpt was introduced. many popular developers , influencers mocked it for not being good and not being able to replace them. but what about 2026. where ai used on bill boards ads, television, news etc.

now when it comes to coding , models like claude sonnet 4.6 or opus 4.7 are super good at coding, it can create millions of lines of code without introducing any syntax issues or bugs, handling exceptions, and super cool front end designs.

now on our day job, we get access to these coding tools, companies are paying 100s of dollars on it to make us productive.

but with the mindset earlier , we where not using the token limits , we are wasting it right ?

So i would like to give you a framework that works for your company to be productive and for you in the long run.

1. Bootstrap the project yourself.

Create the folder structure, write the first middlewares , routes, library or db configurations.

2. Introducing something complex ? write it yourself first then hand over to AI . you have to offload the repeatation part to AI anyway.

For example,

write the first api route, expection handlers, middlewares on the appliacation, hand code it. the authentication logic etc.

like we have an endpoint that lists the tasks

GET /tasks

now you will write this endpoint,

now this single endpoint will return all the tasks as an array of objects right ?

what about listing a particular task ?

GET /tasks/:id will return that right ? SO HAND OVER THIS TO AI, and the deletion route.

you only have to review it,like make sure you are not exposing the fields by using a select query.

now what if someone hammers this endpoint with a 1000 requests, you are quering your DB for the 1000th time for that stupid script kiddy right ?

so you introduce rate limiting,

now you will do the rate limiting part yourself, once , create a re-usable function, and handle the client side part as well. with something like react query, or if you have time, do it yourself with a custom hook. now that part is done.

Now you know how to build things yourself. but you are offloading the 90% of the repetition to AI.

what's the point of hand coding the same stuff for the 100th time anyway ? we choose software engineering because we don't have to repeat ourself like a normal gov employee till the rest of their life right ?

So now we have the application ready in no time and you will have the confidence because you built the foundation. you do not have to repeat it a thousand time because you do not want to memorize the syntax.

now this is the easier bit. what if something is stuck , like a bug that is taking 100 hours to fix ?

Debugging With AI

Ai is super good at reading logs, creating it. finding syntax issues etc. you can absolutely offload the debugging codebase part to AI. no need to spend time on it right ? NO

you can aboslutely use ai for creating logs, reading though big log files, and explaining the issues or auditing it. fixing the syntax can be offloaded to the AI because on your CV you are not going to mention that you fixed 69 syntax issues or type errors.

But what about the bugs that are from a native system or from the os, or you did soemthing wrong and the application crashed in a particular device modal on production ? now , if you are using ai to debug that issue like this,

help me with this , my android application crashed on our users redmi note 5 pro please fix it

that is not going to fix the issue, instead it will introduce more bugs, even if you use the latest and greatest AI modals.

what you actually need is solid proof and a way to re-produce the bugs. so as a software engineer you need error logging, cloud bug reporting , so you use something like sentry or google analytics. you have to implement feature flags, selected distribution, expection handling within the application, if something is failing you should create a log from the application, now these things should be there to fix this particular issue in the first place. because if you had vibe coded the app. you would miss it. anyway, because you would be focused on building new shit features and screens rather than making the product solid. if not. then AI will make you do it, becuase it creates this false confidence of GOD power.

Now real engineering is this. where you care about the existing software on the market rather than introducing features that no one asked for.

Soon or later if you dont focus these then your app/software will fail, no matter how much money you put it in. You can see how bad microsoft windows become in these days when they actually stopped caring about people and introducing silly mistakes like an intern.

Now i picked this example because i think its relevent in here , because microsoft have this huge manpower, the greatest engineering power. so instead of fixing the product they are shipping new shit every month. and making the customers life more and more difficult.

You are not microsoft, but with AI, you will get the same feeling, like the absolute power of shipping new features to make your product standing out.

This for a company , an individual is bad.

Moving forward

As i said you cannot be in the same company for the next decades,

interviews will be the same way like it is in 2020. (pre AI era)

They will ask you library / framework related questions, language specific questions like Math.max() ? or architectural questions, like why even if stateless auth sounds promising, production apps like google use statefull authentications ? what if a hacker gets your admins jwt. how will you invalidate it ? will you wait for it to expire ? or will you take our server down ?

So now it's up to you to make things work well for you, for your career.

WILL YOU BE LIKE A HIGH SCHOOL KID MAKING HIS FIRST WEBSITE WITH CLAUDE CODE ? OR AN ENGINEER THAT THINKS ABOUT SOFTWARE ?

because there is no good metric to know who is a good vibe coder and who is not, anyone who knows basic english can make the ai code. but an engineer will THINK first and then make the AI Code.

The company you are working in doesn't care about your career growth, they don't want you to join another company eigther, so you have to find the time to work on your self. learn new tech. if the job feels repeating or prompting feels like a burden, rethink about it. think globally and not to limit yourself to a small town, a city or with people that has this backward mentality.