Intro
I think LLMs are great tools -- they excel at producing simple API's, , building common ui's or summarizing things. I enjoy using Gemini in writing simple SQL queries for reports I sometimes tend to use copilot to help me to achieve 80% unit test coverage required by enterprise -- I don't hate AI.
But while I do that I also know and firmly believe following things:
- LLM in an IDE is bad for jr engineers its a lot harder to learn fundamentals while relying on LLM's to write code and propose a solutions for them
- LLMs are not good at performance or security (they write a single threaded synchronous code by default, and they do prefer to use whatever has the most mentions(hello react-server components)).
- Using excessive amount of AI generated code is bad for product(looking at you microsoft). I think it comes from Review fatigue primarily its very hard to review all this slop especially knowing that you are not teaching anyone but rather just talking to LLM by proxy.
With all the points mentioned above I used to mentor my team and my engineers accordingly: to use AI when you want but know the downsides and always review your own code.
This morning...This morning I found out that my views differ drastically with my company.
Below is a story from this morning(names are changed) but other that I wrote it as it happen(aka NOTHING is made up)
The story
Zoom opens up and here we are senior leads in a meeting with a VP of 10_000 people+ wanna be software engineering company. Its about 15 of us - we lead teams, we inspire, we mentor, we build things when we have free time from meetings and all of above.
VP Lets call them Kevin starts with a story(he likes stories)
Kevin: Last weekend I was away and I decided to play with AI myself and you know it took me 2 days but I have managed to build a calculator! Not a simple calculator mind you but a complex one(an engineering calculator) and it only took me 2 days with a setup and everything.
Kevin talks about how he knows how to write assembler and c but has no idea about js+html and that's what LLM gave it A CALCULATOR(not a simple one mind you) that runs and works from the box in the browser.
I sit in disbelief....alright --- "calculator" we have millions of calculator tutorials of course LLM can churn it out why is Kevin so proud of it.
Now things take a turn a turn for the worst, Kevin shares his screen and projects a profile of an engineer in windsurf who used it for 99% of his code and says we all have to strive to be him
Kevin keeps going and tells us that skills acquired by knowledge of using LLM's will help our engineer in their careers and if you don't use it - you die (not you literally, you as your career I guess)
Kevin keeps going and asks leads(us) about how are we planning to adopt AI. We start talking how we are planning to stick AI to every single product (no matter if would benefit or be downgrade for it)
Kevin interrupts: No I need to know how your engineers use AI daily, how much do YOU use AI.
People struggle to answer.
Director says: oh I use AI daily and its great and cant stop using it as a tool to tell me what changed in the codebase,
Other lead says: Ai is the smartest engineer I've talked too.
I have to go off camera because I cant help it but giggle hysterically at this people turn bots (yes you are absolutely correct....)
Well this marvelous exchange comes to the end and Kevin gives action points as someone in their position should:
- During StandUps ask what percentage of code written by ai and why is it not 100%?
- We dont simply look for lines of code written but we look into count of log lines written AI
- Challenge every line of the code written not by AI
The Questions
My questions to you -- dev.to community:
Am i a minority to think the things that I've outlined in Intro?
Am I holding my people back by suggesting them to understand the problem and solution before asking LLM to solve it for them?
P.S
Mostly it was just a rant I had to share thank you for reading!
Top comments (1)
You are not alone. I think that finding a solution on your own develops critical thinking. And if the developer does not know available options and does not know how to distinguish them, then LLM will make the choice for him. As my practice shows, there are no bad decisions in development, everything is solved by compromises that LLM is often unaware of.