In an overheard anecdote, Microsoft intern Jinyan Su describes onboarding hindered by teammates' over-reliance on AI agents

@SuJinyan6 This is the AI-native way of working and learning . I think it’s actually more efficient to learn from AI and collaborate with AI. Most senior folks don’t know much compared to AI.

Don’t get me wrong, I am not complaining anything specific to MSFT, I feel much happier and emotionally lighter since I started to work there as it is a place with really nice people and good culture . What I am concerned is how we potentially relying too much on AI (which happens in all industries) that we might be missing the part where we used to get from work— personal growth and learning.

@harry_ai_nerd Is efficiency all we care about?

@SuJinyan6 You can outsource your thinking but not your understanding.

@SuJinyan6 It’s work so we care about efficiency

@SuJinyan6 very common unfortunately, big tech is full of people who joined in the last 2-5 years or so but never really completely ramped up on their team's work

When senior engineer cannot explain what agent does or how the system works it’s exactly the worst possible scenario for humans responsible for things that have to be reliable, secure, safe.

Imagine your doctor that works on the heart doesnt know how to hold a knife or how the heart actually functions. And when agent fails he just helplessly watches you die.

Is that the future you wanna go into?

What you're experiencing possibly is a culture downstream of layoffs and AI use pressure.

I don't know the specifics of MSFT, but where I work there's a lot of pressure within orgs to do things efficiently and deliver more since everyone has access to AI as a productivity tool.

Because you're trying to do a lot more and far more context switching, it gets harder to pay attention to low level details since the volume of code that LLMs generate leads to review fatigue.

@SuJinyan6 Welcome to the new era, code as doc, you don’t really need others to explain things for you now, a pointer to the codebase etc is good enough. The rest is just asking your agent. You can literally onboard anything with your agent, your agent is your mentor

@SuJinyan6 We are all going to lose our ability to code slowly and the all at once. Soon every code base will just become a blackbox which no one understands and we become totally dependent on the agents to get any work done.

@SuJinyan6 It's been pushed really hard, but I don't think most engineers have a good model for when to ask the AI vs when to think for themselves, and certainly not how to teach that. Seeking out personal growth is essential now (e.g. explicit mentoring), I sure wouldn't give that to AI!

@SuJinyan6 😞😞😞 idk where this going towards all you can do is be curious think different and leverage tools be self dependent 💔(find some really good group who can talk and debate with u about research .)

@SuJinyan6 Things are getting worse… I was asking my lab mate some research-related questions, and he told me I should just ask ChatGPT! We used to have long technical conversations with my previous groups and I miss it a lot.

@SuJinyan6 how does one have a phd and can't spell awkward

@SuJinyan6 Well, that was awkward. It’s spelt “awkward” not “ackward” 😝

@SuJinyan6 This is for PhD internship??

@SuJinyan6 It's tough when over-reliance on AI replaces genuine mentorship and understanding. True collaboration should be about learning the "why" behind the code, not just prompt engineering in silence.

@SuJinyan6 the weird part is that some people stopped learning how to ask them

@SuJinyan6 This is the ultimate symptom of Vibe Coding. When engineers offload first-principles thinking to a probabilistic agent, they lose their own mental model. If you can’t debug without an LLM, you aren't an engineer you're just a supervisor for a black box

@SuJinyan6 @harry_ai_nerd Its a company so yeah