Jesus, take the wheel. 🚗
And Github Copilot, take the IDE. 💻
Github says, 92% of US devs are using Copilot.
What. Seriously?!
When did you hear 92% of any demographic using a singular thing?
Unless of course... you say 100% of all people who ever existed, consumed di-hydrogen monoxide.
(There's exactly one way this line gets dark. Don't go there. 👀)
Join me on a quick journey as I talk about:
- What my fears actually are.
- How other experienced devs feel about this.
- Maybe I'm worried about nothing?
- And how WE are responsible for how we use LLMs!
🔥 When the machines took over the world in 2024
After a quick Google search, it seems like most devs are using AI assistance in writing code. I’d be lying if I said I haven’t used AI to write code at all. Of course, I have. I don’t live under a rock.
I've seen devs get oddly comfortable with the idea of sharing their code related data with third party cloud services, which are often not SOC2 (or something similar) certified, and make vague unprovable claims of privacy at best.
Things like Github Copilot (and Copilot chat), Bito.ai, and several other AI code extensions on the VS Code marketplace have more than 30 million installs. Crazy! 🤯
And then there's me. I’ve not made AI assistance a part of my regular code workflow. A couple of times I’ve taken help from GPT to get some boilerplate written, sure. But those times are an exception. Things like Github Copilot, or any kind of code review, code generation tool, PR creation, or commit-assistance, isn’t a part of my IDE or CLI flow.
Maybe it’ll change with time. We’ll see.
“But… why?”
😟 What I'm ACTUALLY worried about
The answer is simple. 👇
1. I fear my programming skills will get rusty
I am concerned that the way I write and read code will suffer if I get too used to AI assistance.
- I’m concerned I’ll begin to overlook flaws in code that I can catch otherwise.
- That I’ll start to take AI generated code for granted.
- And looking up APIs, built-in methods, or other documentation will start to feel like a chore.
I fear… that I’ll start slipping.
2. I'm not comfortable enough with sharing all of my code with a third party service
Companies can be damn smart about inferring things from the data you provide. Sometimes they'll know about things that your family won't know about.
The idea that sensitive business logic may get leaked to a third party service, which may eventually be used to make inferences I'm not comfortable with or just... straight-up leak? I mean, software gets hacked all the time.
I think I'm being very reasonable thinking that I don't want to expose something as sensitive as code in an unrestricted manner to a third party company. Even if that company is Microsoft, because even they f*ck up.
👀 From the Experienced Devs' point of view
This isn’t a take that is unique to me either!
1. More-experienced devs tend to not want to lean on “crutches” to write their code.
I’ve even had the pleasure to work with senior devs who didn’t want to use colored themes on their IDEs because they thought it’ll hurt their ability to scan, read, or debug code! (that was a bit much for me too)
After all, “programming skills” is a lot more than just writing code.
2. Older devs have seen all kinds of software get hacked, data leaked, etc.
I mean, when haveibeenpwned.com sends you emails about your credentials, emails, and other data leaks every year for over 10 years... MANY TIMES from billion dollar corporations...
When you hear "When you're not paying for the product, you are the product" for the bazillionth time, which is then backed by yet another company selling that data to some third party...
Yeah... it gets tiring.
And it gets real easy to just disconnect as many wires as you can and go back to stone age.
“Older devs”? Am I… Am I getting old?
Nah, I’m still 22 and this is 2016 or something… right? Right??
Btw, the answer to the question in the title is 👆 this. Congrats! The post is over! Over to the next one…
Buuuuut… if you want to continue reading…
🚶 Let's take a step back for a moment...
I think my fears may be exaggerated.
Let's keep the whole data privacy angle aside for now, because that's a whole other topic on it's own that I feel about rather passionately.
I personally don’t have enough data to empirically say that using AI assistance will bring about the doom I fear… That it’ll downgrade me from what I am today, to an SDE1.
But I’ve seen patterns.
- I’ve seen AI-generated sub-par quality code go through code reviews and end up on the
mainbranch. - I’ve seen library functions being used without properly understanding what, or what alternatives exist just because an LLM generated that.
- I’ve even seen code generated to solve a problem, for which a utility function already existed in the codebase but wasn’t used because knowing this utility existed was a lot more work than asking GPT to generate it for you.
💎 Diamonds are Bad code is forever
“Wait a damn minute… I’ve seen this movie before!”
- LLMs are a pretty new thing… but 💩 code has been eternal!
- Every. Single. Dev. Ever. Has used library functions without fully understanding it or looking at alternatives. You, and me, are both guilty of that. (What? You thought
Array.prototype.sortwas the best way to sort anything? It’s just sufficient in most cases!) - A piece of logic gets reinvented (re-copy-pasted) all the damn time! Just that before it used to be from StackOverflow, now it’s from ChatGPT.
🤷 So, what’s the big fuss about?
"Will using ChatGPT make me a bad programmer?"
I think, no.
The takeaway is that you just need to care about what you build.
Take pride in what you build.
🤖 Where the heck does LLM/AI fit in?
LLMs are not inherently evil.
In fact, they can be pretty damn useful if used responsibly:
- Quality Code: An LLM might handle edge-cases that a less diligent developer wouldn’t consider.
- Comprehensive Tests: LLMs might write tests that are more comprehensive than what some devs would produce.
- Comprehensive Types: It might even write types more "completely" than an average dev might write on their own, or might have the skill to write.
However, the responsibility lies with the developer to ensure that the code output is guarded and well-monitored. Someone who doesn’t care would have done a shoddy job at any point in history. The presence of LLMs doesn’t change that.
😎 The art of actually giving a f*ck
There's a lot of devs out there who don't care.
But you’re no such dev. You DO care.
Else you wouldn’t be here on dev.to learning from people’s experiences.
I recently wrote about what new devs should care about to grow in their career. It’s a LOT MORE than just code.
Maybe I’ll introduce some AI in my VSCode.
I think it’s a matter of when, instead of if.
What’s more important is… as long as I care about making sure my programming output is readable, performant, high quality, and easily reviewable, I think I’ll be fine, and so will you.
👇 P.S.
If you want an example of something I care deeply about, and has both great code 💪 and… less than excellent code 🤣, take a look at our open-source repo!
It’s something that lets you spot how long it takes for you to deliver your code, how many times PRs get stuck in a review loop, and just generally how great your team ships code.
middlewarehq
/
middleware
✨ Open-source dev productivity platform for engineering teams ✨
Open-source engineering management that unlocks developer potential
Introduction
Middleware is an open-source tool designed to help engineering leaders measure and analyze the effectiveness of their teams using the DORA metrics. The DORA metrics are a set of four key values that provide insights into software delivery performance and operational efficiency.
They are:
- Deployment Frequency: The frequency of code deployments to production or an operational environment.
- Lead Time for Changes: The time it takes for a commit to make it into production.
- Mean Time to Restore: The time it takes to restore service after an incident or failure.
- Change Failure Rate: The percentage of deployments that result in failures or require remediation.
Table of Contents
Top comments (18)
This article is an antithesis of what a real developer should be. In first instance, take leverage of AI and learn from it. However you should avoid taking for granted what AI produces.
You should take documentation in first place and from there build up.
Personally, I don't feel threatened by AI as I take decisions based on human experience, while AI only produces content as reflection of past experiences.
Should a dev be afraid of AI? Not at all.
Good take, Erick.
Devs just need to be responsible with introducing AI tools in their workflows. Starting small and seeing how it works, what it's good at and what it's not.
Though I am iffy on using terms like "real developer".
I slightly disagree, I think devs should commit mistakes with AI and then maybe learn from it. But ofc mistakes could be costly for orgs. That's where PR reviews, senior expertise come in right?!
Devs should definitely not avoid AI of course.
But I think maybe starting with a production-grade codebase isn't the best idea, regardless of whether senior devs are involved or not.
Devs should spend enough time learning how to use AI as a tool responsibly in their own personal projects, get community, friends, and coworkers feedback on it before involving AI in production grade code.
I am also in those 8% of devs not using co-pilot and my main concerns are the same as the ones mentioned here. But I still use AI every day at work to make my life easier. LLMs can boost your productivity to 10x if you are able to understand the generated solutions and iterate over those to make them better. Great Read :)
The trick is to not copy/paste production secrets into this thing. 🤣👇
Such a divisive topic. Clearly there are pros and cons and both are valid actually.
I would say that having an AI generating most of your code is not a good thing. And I would also say that not incorporating an AI would be a lost opportunity. So my take is to use it simply as a pair programmer. I'm driving, the AI might suggest some things I don't agree with - and I readily ignore those. Other times I might look at the code I've written and ask the AI if there is a simpler way to do something and it, again, may come up with something valuable, or some utter nonsense that I don't want in my codebase.
I think there is real risk in letting it write your code for you, but I see real value in using it as a tool to potentially improve parts of your work while you're doing it. It's great to pair program, now we all have access to one. Just use the tool appropriately, because all it is, is a tool.
the competences of new code learners getting weaker day by day,
something like patience, searching for month about issue you faced while you code, watching course of 100 Videos, Or crash course of 10 Hours (Crash :) )
all these wonderful moments they wouldn't live it.
I speak from my experience with my classmates and i feel that the harm of AI for beginner is stronger than for professionals.
yes they won time and submit their assignments in the best way but they lose skills through two year of studying.
AI became an alternative mind think instead of them.
I'm a self-taught MERN stack developer. Without ChatGPT and Copilot, I wouldn't have been able to release a SaaS platform that's now generating some revenue for my startup. As a solo founder and developer, AI helps me write 96% of my code. I'm very comfortable with this approach because I review and correct the AI-generated code line by line before using it. I've become extremely efficient at prompting AI, and there's no going back for me. When I start hiring developers, those who don't use AI will be let go. I'm serious about this! 😄
Beautifully stated. As a mern developer with years of experience before ChatGPT and Co Pilot. There tools are priceless in regards to the speed and assistance they provide when writing code.
Relying solely on AI for all your coding needs risks undermining your development as a programmer, potentially reducing you to a copy-and-paste operator rather than a thoughtful, creative coder. Personal coding style and logic are crucial for fostering innovation and problem-solving skills.
Fear surrounding the current state of AI is largely unfounded. Language models and generative models are simply tools, and being skeptical of them is akin to using a rock to drive in a nail when you have a hammer available.
In reality, AI and language models are not inherently intelligent. Without creative input and context, they are essentially empty shells. This means that if you are not already skilled at something, AI is unlikely to make you significantly better.
However, AI and language models can be incredibly useful in augmenting our existing skills. For instance, I love writing code, but I often struggle with writing clear and concise documentation and comments. Here, AI and language models have been incredibly helpful. By providing the model with a rough idea of what I want to say, it can generate a draft that I can then refine and edit. This can save me a significant amount of time and mental energy, freeing me up to focus on other creative tasks.
In my view, the development of AI and language models should be encouraged, rather than stifled. By improving these tools, we can achieve more. As with any technology, there are certainly risks and challenges to be addressed, but I believe that the potential benefits far outweigh the drawbacks.
In short, AI and language models are not a solution to any of our problems, but they can be incredibly useful tools when used in the right way, which is augmenting current skills and creative pursuits. By embracing these technologies and working to improve them, we can do more.
I briefly used Copilot and then stopped, I prefer Tabnine. Copilot gets in my way often, but I do use AI often for boilerplate code or research. It's a tool like any other tool.
If you approach AI as a tool, to help you speed up your workflow then you'll be okay, but if you just copy and paste code you don't understand just to complete the task, then in 6 months or a year you'll be in trouble.
A developer's main purpose is to analyze problems and find solutions, how you connect the dots is up to you and your team, as long as the solution results in code that is maintainable, and takes into account various factors such as scale, security, and so forth.
As for the security and privacy issue, just be aware anything on the internet can be leaked or hacked, even code. Microsoft, OpenAI, or any other vendor for that matter will have mass lawsuits if they breach their SLAs. So I don't think sharing code with these services is a major problem.
Sure these companies will use your code to train their models, it's anonymized as tokens, similar to telemetry. Of course follow sensible good practices, like not putting important sensitive keys, passwords, and documents in your codebase. Inject these at runtime in the environment via ENVs or behind APIs.
as a friend of mine say
"you didn't know how to use stackoverflow to solve your problem why do you think using AI will do?"
I've already read Mark Manson's book a few times to develop the IDGAF muscle for cases like these.
I use AI because it's way less mental struggle to correct bad code than it is to write code from scratch. ADHD is funny that way.
Imho, using AI tools extensively as a developer changes you, and not in a good way.
I used them for a while. Meh. 🫤
I prefer: think yourself, look up yourself. Stay sharp.
I use ChatGPT.4 as my help at other hand but I don't like GitHub Copilot and I think people should write code on their own not like "Write an alternative to Finder on Python".
Some comments may only be visible to logged-in visitors. Sign in to view all comments. Some comments have been hidden by the post's author - find out more