Well some introduction is needed.

Intro to view

It’s little confusing now when I start to write this.
It sounded like a great idea 5 minutes back .
To me View is a post which provides my opinion on a topic being the in the well.

Why in a well ?

To give you some insight into why I have a weird view of a you might need some history about me.
Not to make it too long, I make money by making software or at least I used to.
Now I kind of do blueprints of software and let other smarter people build it (sometimes I build some parts of my own blueprint).
I have been doing it around 9 years now (which feels weird to say ) and that too in the same company.
Over the years, my surroundings have become my well.
I do hear and read croaking outside the well but I can explain what I see in this towering well where I exist the best.
Over the years, all the ribbits I hear and read have been sitting in my head wanting to escape.
So here I’m writing about it to give them the escape.
Hope it gives you some insight into these ribbitty articles.

LLMs: auto correct on steroids or just self mimicking parrot

Well this might be career ending if some corporate recruiter or AI influencer reads it .
Most people just call it AI now even though it’s no where close to it.
LLMs have been a big revolution across most industry or at least it is painted as such.
It’s hard to deny the facts from the past versions I have seen in 2021 to now they have improved significantly.
So in most honorable way here is my opinion.

Experience

Past

It was being pushed around a lot by everyone in management (mostly due to KPI and my company’s indirect cash flow from it).
In my organization, I was one of the first few beta testers for Github Copilot.
Around 2021, trying to include LLMs in my dev workflow felt trash.
I couldn’t see the value add.
Auto complete felt annoying and mostly useless.
Most boilerplate generated I could do with most general IDE features.
It felt like a fluke and ok to forgot.

Change

Around end of last year, I could see a shift working with devs who were really progressing fast with it.
It felt like it needed a retest.
ChatGPT didn’t feel like the best fit but Claude felt interesting.
I had tried it outside the copilot subscription provided by my company.
It seemed interesting doing small scripts on my own and seeing Claude generate a very close versions of it was definitely interesting.
I felt like I was lacking a better way of approaching it as constantly prompting felt weird and slow.
Seeing some stuff online and seeing full agent mode was quite intriguing.
As the blueprint guy, trying quick POCs or doing things I didn’t care about this way at office felt time saving.
Recently, they introduced Claude Opus and Sonnet in copilot.
With Agent CLIs, things felt faster with instructions, pre-defined prompts you could run some small projects without much effort.

Corpo Experience

Inside office, Always being on time crunch letting Claude manage souless crap with some markdown files felt great.
I could focus on more important things like getting out of office on time.
To be frank, in my my organization nobody kind of cared about code in general.
All these years of working hard to getting good at it felt weird.
As much I was good at it and wanted to showcase it, I could see the stare down from management treating me like a kid and nerd who were letting me play for their own goals.
Experience was always soul crushing always leaving the very few people who cared, most took it as a means to an end.
A part of me died with this and letting an LLM do the part to be honest felt ok in this environment (more like I don’t care anymore about what I do here. They don’t care if it’s maintainable or just slop as long as it works).
While things continued, the muscle memory of many years tapping on a keyboard still persists.
Major caveat that I saw was mostly for new engineers who were all in on prompting.
To add a little history my company is in a very traditional sector not a big tech, the quality of hire is generally very below averages we just get horrible contractors or good ones who leave within a year.
How I have stuck around is still a mystery (maybe I’m bad at my job).
Coming back to it, I could see very clearly they couldn’t explain what they were doing or what they had created.
Debugging was a nightmare for them.
They didn’t understand a single line of code and were treating it like a black box.
With a lot of legacy systems and libraries experience for these people was worst.
For simple tasks, there were done in less than a day but for even slightly more complexity multiple sprints would be burnt down not delivering anything.
I have seen the recent thing with Amazon getting all hands on deck about all PRs from mid-junior devs needing senior approval.
Still I feel it’s going back to the Watchman problem of software development.
With all the flood of PRs with very few Senior Devs it’s already a nightmare.
All senior devs are the watchmen of their codebases but who watches to watchmen when the fuck up which they’ll with limited bandwidth and a flood of work with no effort being pushed to them.
All junior-mid level devs become worse at their jobs and never ideally becoming senior devs who can actually review the code as they have never written or reviewed anything.
The cycle is crazy.

Personal Experience

At home with all personal and side projects strict rule is to avoid LLMs at all cost.
Primarily, building apps is a fun thing for me that I got enchanted since I was a kid.
With projects which are close to me, I take pride in writing every line myself and getting the dopamine hit seeing it work.
It makes me feel my energy is actually going to things I really care about.
Spending 17$/200$ per month feels too much for me at home for something I can do myself.
Now my skills feel like magic and myself as .
Burning down a forest and drying up rivers for my shitty personal project is not worth it.
I don’t need to scrap all of github to write multiple lines of shitty code.
I can create shitty code from my mind at home.
So to say this whole shit app is written from ground up by a person who gives you his view from the eyes of a .

Conclusions

• Let LLM do souless crap it’s made for it particularly at work
• If you are a senior dev, I feel for you reviewing all the slop from bitches around you
• Spend true focus on things you like and do it yourself
• LLMs are just tools which make you think you are smart but whether you are dumb or smart they are making you dumb (if you are nothing without the suit. you don’t deserve it)
• Fuck you Sam Altman, Nvidia and MicroSlop, you can’t take this from me I’m a of code