AI Writes My Code Now. So What Am I For? | An ML Engineer’s Honest Take
https://www.youtube.com/watch?v=wEcpzfWgxOE
无作者
When AI writes your code and it feels like nothing
Lee,have you ever wrote down the email, or wrote down the code and pushed it and felt like nothing, because AI wrote most of it and you just reviewed and approved it.I am a machine learning engineer, and that feeling has been following me for months.Then one day it got even worse when I was developing and researching,reading paper and developing the models and [snorts] testing architecture,my manager would walk over to me and said, “Oh,I just put it in chat GPT and it worked perfectly.”, And I didn’t know what to say because he was right.
Why I’ve been away: AI depression
I’ve not been posting for a while.It started with a personal events in my life,but it kind of spiraled to kind of AI depression.So, if you’re a developer who’s feeling anxious,lost, or quietly wondering whether you are still needed,this video is for you.I’ll be honest, that what it feels like inside my job as a machine learning engineer, right now, and share what I’ve learned it so far.I don’t have it all figured out,but but I’d rather think it through with you guys than I pretend, I know everything.So, as its core,
The disappearing joy of writing code
I think an engineer’s key role is to plan, define the problem,plan it out how to solve this issue in a technical way,and write the code because our way of solving problem is writing a code.But as you all know, writing code part has been disappearing.Most engineers, software engineers,any data engineers, or machine learning engineers would know.We write far less code by hand than we did two, three years ago.AI writes the first draft and we review it and then we push it and merge it.And honestly, the writing code part,it is stressful.The errors chase us hour by hour.Another error leads to another error.Digging through the documentation, and going through our beloved Stack Overflow post,but that was also part of joy.Believe it or not, that was part of joy of our engineers.In that moment, the bug finally disappears,like poof.And every day that was kind of a milestone, of our personal and daily achievements in our job.So, when that goes,the question creeps in.So, what’s my value now?
What’s my value now as an engineer?Am I being obsolete?Am I going to be replaced by AI in the end?
You’re not alone
So, when I went to looking online,I realized how many of us feel this way.One developer in Reddit wrote that they don’t feel like developer anymore.They felt they’d almost forgotten how to code,and the replies were full of people who agree on it.And another AI engineer posted that because AI models are so good,they feel irrelevant.Someone replied that a task that had they have been planning for 5 days, was done in a single prompt.If you felt that way, you’re not being dramatic.We all are.You’re noticing something real that is happening.But people are not really covering that much.
Mixed signals: use AI at work, code by hand in interviews
And the thing that is really confusing nowadays is about the mixed signals.At work, we are told to use AI for everything.Some company use cloud, some company use ChatGPT or Microsoft Copilot.Everyone have their own AI tool in their company.But many interviews, technical interviews,still include the stage of writing code by hand.So, meanwhile,the new AI models come out every few months,each one more capable than one another.
How the machine learning field is changing
For me, we can zoom into how the machine learning field is transitioning.It hits in a specific way.When I first started studying machine learning,and as I’ve shared on the machine learning road map,the core skill was mostly about building models,understanding what are the machine learning models out there,in what occasion we should use it, how to adjust it,and how to improve it.So, if you look at Kaggle,the data science competition website was full of challenges,like let’s detect the breast cancer based on the images,or let’s detect fraudulent transactions looking at the numeric data.So we learned scikit-learn and PyTorch inside out.That was the one of the most, important skill that you could, find anywhere in the job description.That was the job.Now a lot of time I’m competing with a tool AI tool that anyone can use.The job this AI related job, machine learning related job is keep expanding.Many people have been asking me in my work or my friends were asking me “Oh,
can you make AI agents for me?Can we run our own ChatGPT in our servers?” Like local large language model.Suddenly everyone expect us machine learning engineers, to be expert in the things that barely existed a year ago.And I think what I found interest find interesting is if you go to LinkedIn,there are many people who are suddenly expert in this AI agent within a year.So where does this leave us?
So where does this leave us?
I’ll be honest, I don’t have a neat answer because AI is changing day by day.Not using AI however isn’t really an option anymore.I use it every day,but after a lot of thinking I’ve come to believe engineers still matter.Not because we’re faster than AI, we’re not and we can’t,but because there are a few things I think we still do better.
1 Communication
The first is communication.Even if AI writes the code and opens the pull request,I’m still the one who reports back to our manager, right?I’m the one who has to explain what the real problem was and why we chose, the solution, and how it is doing.Say an API endpoint is slow.And AI fixes by adding a cache, the test pass and the pull request looks clear.Then my manager would ask, “Why did we add caching here?Could customers see outdated prices?” If my only answer is,“That’s what Claude did.Let me go and ask my Claude and see if it remembers my context.That’s not an answer.That’s not a professional, answer that any manager or any directors above us would expect.So, I’ve started interrogating AI as the way my manager would ask questions.Why caching and not a faster database query?How long does the data stay stale?What happens if it goes wrong?I can’t answer those myself, then I’m not ready to hand over my work.So, you are the key person who mitigates the AI the work done by AI,
but to your manager to the decision-making process.
2 Responsibility and ownership
So, the second is responsibility and for me that means ownership.AI can finish 10 tasks, let’s say,30 minutes, maybe sometimes 10 minutes.But that doesn’t mean I should move and on that speed.It is that reality that many companies, sometimes expect you to solely be AI human,but we don’t have AI brain on us.We can’t work on AI speed.Well, we need to know how to work on our own pace in collaboration with AI.So, let’s picture this.You give AI agent like 10 bug-related, task and half an hour later they’re all closed.But one of those fixes quietly changed a shared function, and something breaks in a completely different part of the app.So, nobody noticed for a week.When it comes back to you, you can’t even remember reviewing it.This happens.So, I’ve started doing the opposite of what AI speed suggests.I only ask AI to do as much as my brain can actually understand and review.Maybe that’s can be a three task among 10 of them.It sounds slower, but it’s what keeps me in control.
Using AI at my own tempo is how I stay the owner of the work,not just a person who pressed a button.And also that’s the thing that makes me irrepla-, irreplaceable.
3 Curiosity and critical thinking
And the third is curiosity and critical thinking.For a while,I honestly wondered whether we still need to learn the fundamentals at all,like writing a code, machine learning basics,the theories.That was also the question that really, chased me because I’m a content creator who talks about how to learn AI, and also share my experience of how to learn AI.This is the part that confused me a lot and also make made me, fall into this AI depression.The conclusion that I reached is yes, we do.We still need to learn the fundamentals, including learning how to program.So, here’s a classic example for machine learning.
Example: the 99% accuracy trap
You ask AI to build a fraud detection model, and it probably repeat reports back to me.It has 99% accuracy.It sounds amazing, but if only 1% transactions are fraud,a model that says not fraud to every single one also gets 99%.It hasn’t caught anything.You only catch that if you know the basics.When the data is unbalanced, accuracy can lie.So, you can ask how many frauds it actually caught.AI didn’t flag the problem unless I asked the questions.Knowing the fundamentals led me to.And every time I ask “Wait, why did you do it like this way?Why did you approach it this way?”, I don’t just get a better result by asking back the questions,I learned something, too.I learned something that AI knows that I don’t know or I have overlooked.That’s how AI can make us sharper, not lazier.So, I’m still working on all of this.
Some days are still hard
Some days I still feel this AI depression, just swallowing me like the giant wave coming towards me.But if you’ve been feeling like you’re not a real developer anymore,like I sometimes feel, and it’s absolutely all right to feel that way.I want you to know that you’re not alone.Every single one of us is feeling this way.My friends say, “Oh,you’re a machine learning engineer,your job must be safe from the the AI.” Maybe, um,but I think because majority of our work is also following up on our research, and also writing code,we are also in the risk of getting replaced.But, these three,communication, responsibility,and critical thinking are things that I think, that would make us irreplaceable and keep us alerted, to be a thinking human being.What I want to say is those who feel anxious, and feel like you’re not worth it anymore,
Your turn: do engineers still matter?
thinking that way and feeling feeling that you’re falling behind doesn’t mean, it doesn’t mean you’re not doing well.It means you care about your craft and it is meaningful, by the fact that you’re thinking about, how your skill would proceed.So,I like to really hear from you what part of your work do you miss doing, by your hand?And do you think engineers still matter?Or am I being too hopeful about it?Tell me in the comments.I try to read all all of them and see you next time.