Forward-thinking companies should use AI to transform each developer into a “10x developer,”
Developer + AI ≠ Developer x 10
At best, it means 1.25 x Developer, but in most cases, it will mean 0.5 x Developer. Because AI cannot be trusted to generate safe, reliable code.
I think 10x is a reasonable long term goal, given continued improvements in models, agentic systems, tooling, and proper use of them.
It’s close already for some use cases, for example understanding a new code base with the help of cursor agent is kind of insane.
We’ve only had these tools for a few years, and I expect software development will be unrecognizable in ten more.
It also depends on the usecase. It likely can help you better at throwing webpages together from zero, but will fall apart once it has to be used to generate code for lesser-discussed things. Someone once tried to solve an OpenGL issue I had with ChatGPT, and first it tried to suggest me using SDL2 or GLFW instead, then it spat out a barely working code that was the same as mine, and still wrong.
A lot of it instead (from what I’ve heard from industry connections) being that the employees are being forced to use AI so hard they’re threatened with firings, so they use most of their tokens to amuse themselves with stuff like rewriting the documentation in a pirate style or Old English. And at the very worst, they’re actually working in constant overtime now, because people were fired, contracts were not extended, etc.
It’s made me a 10x developer.
As someone who transitioned form Junior to Dev as we embraced LLMs. Our company saved that much time that we all got a pay rise with a reduction in hours to boot.
Sick of all this anti LLM rhetoric when it’s a tool to aid you. People out here thinking we just ask ChatGPT and copy and paste. Which isn’t the case at all.
It helps you understand topics much quicker, can review code, read documentation, etc.
My boss is the smartest person I’ve ever met in my life and has an insane cv in the dev and open source world. If he is happy to integrate it in our work then I’m fine with it. After all we run a highly successful business with many high profile clients.
Edit: love the downvotes that don’t explain themselves. Like I’m not earning more money for doing less hours and productivity has increased. Feel like many of the haters of LLMs don’t even work in the bloody industry. 😂
Developersdevelopersdevelopersdevelopers,developersdevelopersdevelopersdevelopersAIThis assumes it is about output. 20 years of experience tell me it’s not about output, but about profits and those can be increased without touching output at all. 🤷♂️
*specifically short-term profits. Executives only care about the next quarter and their own incentives/bonuses. Sure the company is eventually hollowed out and left as a wreck, but by then, the C Suite has moved on to their next host org. Rinse and repeat.
Often they only want the illusion of output, just enough to keep the profits eternally rising.
I don’t honestly believe that AI can save me time as a developer. I’ve tried several AI agents and every single one cost me time. I had to hold its hand while it fumbled around the code base, then fix whatever it eventually broke.
I’d imagine companies using AI will need to hire more developers to undo all the damage the AI does to their code base.
AI can absolutely save you time, if you use it right. Don’t expect it to magically be as good as a real programmer… but for instance I made an HTML visualisation of some stuff using Claude, and while it got it a bit wrong, fixing it took me maybe 20 minutes, while writing it from scratch would have taken me at least a couple of hours.
I guess for some simple stuff it can work fine, but the majority of the code I write is not at all simple, and it’s all highly dependent on the libraries I’ve written, which the AI is really bad at learning.
And then in terms of documentation, it is just hopelessly inept.
I’ve found it can just about be useful for “Here’s my data - make a schema of it” or “Here’s my function - make an argparse interface”. Stuff I could do myself but find very tedious. Then I check it, fix its various dumb assumptions, and go from there.
Mostly though it’s like working with an over-presumptuous junior. “Oh no, don’t do that, it’s a bad idea because security! What if (scenario that doesn’t apply)” (when doing something in a sandbox because the secured production bits aren’t yet online and I need to get some work done while IT fanny about fixing things for people that aren’t me).
Something I’ve found it useful for is as a natural language interface for queries that I don’t have the terminology for. As in “I’ve heard of this thing - give me an overview of what the library does?” or “I have this problem - what are popular solutions to it?”. Things where I only know one way to do it and it feels like there’s probably lots of other ways to accomplish it. I might well reject those, but it’s good to know what else exists.
In an ideal world that information would be more readily available elsewhere but search engines are such a bin fire these days.
I was in the same boat about…3mos ago. But recent tooling is kind of making me rethink things. And to be honest I’m kind of surprised. I’m fairly anti-AI.
Is it perfect? Fuck no. But with the right prompts and gates, I’m genuinely surprised. Yes, I still have to tweak, but we’re talking entire features being 80% stubbed in sub 1 minute. More if I want it to test and iterate.
My major concern is the people doing this and not reviewing the code and shipping it. Because it definitely needs massaging…ESPECIALLY for security reasons.
Which tools are you finding success with?
I mostly use AI as advanced autocomplete. But even just using it for documentation is wrong so often that I do’t use it for anything more complex than tutorial level.
I got pretty far with cursor.com when doing basic stuff that i have to spend more time looking up documentation than writing code, but I wouldn’t trust it with complex usec cases at this point.
I check back every 6 months or so, to keep track of the progress. Maybe I can spent my days as a software developer drinking cocktails by the pool yelling prompts into the machine soon, but so far I am not concerned I’ll be replaced anytime soon.
Genuinely a bit shocked to see the number of robolovers in these comments. Very weird, very disheartening. No wonder so much shit online doesn’t work properly lol
Ironically, processing large amounts of data and making soft decisions and planning based on such data makes AI ideal for replacing C-suite members.
Not to mention the cost savings difference. Developer salaries make a ChatGPT subscription look like a bargain. C-level salaries make racks of dedicated hardware to run local models look like one.
Let’s make a community powered, open source project to do this and watch them squirm when investors demand that million dollar CEOs get replaced with AI for higher investor returns.
Pointing this out in company wide meetings is a fun past time.
My theory is that C-suites are actually using “AI efficiency gain” as an excuse for laying off workers without scaring the shareholders.
“I didn’t lay off 10% of the workforce because the company is failing. It’s because… uhmmmm… AI! I have replaced them with AI! Please give us more money.”
It’s the next RTO.
The funny thing is that if AI coding were that good, we would already see widespread adoption in open source projects. But we haven’t, because it sucks. Of course commercial software development companies are free to lie about how much they use AI, or get creative with their metrics so they can get their KPI bonuses. So we can’t really believe anything they say. But we can believe in transparency.
As always, there are so many people selling snake oil by saying the word AI without actually telling you what they mean. Quite obviously there are a great many tools that one could call AI that can be and are and have been used to help do a ton of things, with many of those technologies going back decades. That’s different from using ChatGPT to write your project. Whenever you hear someone write about AI and not give clear definitions, there’s a good chance they’re full of s***.
You can fucking swear on the internet
it means more ambitious, higher-quality products
No … the opposite actually.
Read the article before commenting.
The literal entire thesis is that AI should maintain developer headcounts and just let them be more productive, not reduce headcount in favour of AI.
The irony is that you’re putting in less effort and critical thought into your comment than an AI would.
For the sake of benefit of the doubt, it’s possible to simultaneously understand the thesis of the article, and to hold the opinion that AI doesn’t lead to higher-quality products. That would likely involve agreeing with the premise that laying off workers is a bad idea, but disagreeing (at least partially) with the reasoning why it’s a bad idea.
I get what you’re saying, but the problem is that AI seems to need way more hand holding and double checking before it can be considered ready for deployment.
I’ve used copilot for Ansible/Terraform code and 40-50% of the time it’s just… wrong. It looks right, but it won’t actually function.
For easy, entry programs it’s fine, but I wouldn’t (and don’t) let it near complex projects.
I’ve seen similar issues with ansible and terraform. It’s much better with more traditional languages though. Works great with core go-lang, Python, Java, Kotlin, etc. Ymmv when it comes to some libraries as well. I think it’s mostly to do with the amount of training data.
Its not about writing easy entry programs, it’s about writing code robustly.
Writing out test code where tests are isolated from each other, cover every edge case, and test every line of code, is tedious but pays dividends. AI makes it far less tedious to write out that test code and practice proper test driven development.
A well run dev team with enough senior people that manages the change properly should increase in velocity if they’re already writing robust code, and increase in code quality if they’re not.
AI makes it far less tedious to write out that test code […]
Completely disagree.
In my experience, LLMs constantly generate bad code that needs to be thoroughly checked, to the point that writing by hand is more practical.We use copilot literally every day and it’s extremely helpful, literally not a single developer at our company disagreed on the most recent adoption survey.
Maybe you’re trying to use it to do too much, or in the wrong way?
That middle graph is absolute fucking bullshit. AI is not fucking ever going to replace 75% of developers or I’ve been working way too fucking hard for way to little pay these past 30 years. It might let you cut staff 5-10% because it enables folks to accomplish certain things a bit faster.
Christ on a fucking crutch. Ask developers who are currently using AI (not the ones working for AI companies) how much time and effort it actually saves them. They will tell you.
I use it here and there. it just seems to shift effort from writing code to reading and fixing code. the “amount” of work is about the same.
I hear that. Given I need practice in refactoring code to improve my skills, it’s not useless to me right now but overall it doesn’t seem like a net gain.
It doesn’t have to make sense or make the outcome be better, the only thing it has to do is make the company look better on paper to its shareholders. If something can make the company look better on paper it will be done, the quality of the work is not relevant
Not only the shareholders. If some of the higher level administration can get richer in the short run, even if that might actually hurt the shareholders in the medium run, you can bet that many of them will do so.
I use it so much. All my Google searches for syntax or snippets? Web searches are unuseable at this point, AI can spit it out faster. But the real savings? Repetitive code. I suck at it, I always make typos and it’s draining. I just toss in a table or an api response and tell it what I want and boom
It probably does write 75% of my code by lines, but maybe 5% of the business logic is AI (sometimes I just let it take a crack at a problem, but usually if I have to type it out I might as well code it)
What it’s good at drains my concentration, so doing the grunt work for me is a real force multiplier. I don’t even use it every day, but it might be a 3x multiplier for me and could improve
But here’s the thing - programmers are not replaceable. Not by other humans, not by AI - you learn hyper specific things about what you work on
But the real savings? Repetitive code. I suck at it, I always make typos and it’s draining. I just toss in a table or an api response and tell it what I want and boom
Get better at it, manually, or you’ll suck at it forever. It’s a skill like anything else.
I agree, but I acknowledge we could be at a “cursive writing” moment where something that was once a critical skill becomes irrelevant. That’s sort of a pending question at this point.
I mean I’ve spent a lot of time writing regex to automate large sets of changes. Sometimes it can be a bit fiddly to get the regex just so. Like replacing direct field access with getters where you have to find the field access and change .foo to .getFoo() and the capitalization can take a couple of tries to get just right.
With AI you can literally just say “replace all direct field access (e.g. thing.foo) with getters and setters” and the AI will do it in under a second. It will still be a very useful skill to be able to do things like that with regex because not everything is so easy to communicate to the AI, but it will become less frequently needed and a lot of developers who never learned that skill will get by using AI and just doing the rare things AI can’t do with repetitive keyboarding.
I see cursive writing brought up a lot in these conversations and I don’t think it applies. Firstly, the cognitive load of writing code is higher than writing your letters so they join up. You’re not just making sure you write the letters correctly, you’re also following the syntax rules of the language you’re writing. And while you’re writing, you’re reinforcing those rules in your head. Yes, initially it’s hard and boring.
And yeah, sometimes you get it wrong or forget to capitalize. That is a feature, not a bug. The more you do it, the easier it gets. I spent a couple weeks trying to use CoPilot and at the end I still had to correct its shitty code, which either hallucinated features I wasn’t implementing, or hallucinated syntax rules I wasn’t using. It was like spending a sprint trying to get a subpar intern up to speed. At the end of those two weeks, my manual coding accuracy took a noticeable hit.
I complained to higher-ups and they told me “oh it’s definitely a skill getting the prompt written correctly”, which was patronizing and irritating. Would I rather spend time getting good at asking the proprietary magic thinky box to maybe write good code this time, or would I rather get better at coding?
I mean I’ve spent a lot of time writing regex to automate large sets of changes. Sometimes it can be a bit fiddly to get the regex just so. Like replacing direct field access with > getters where you have to find the field access and change .foo to .getFoo() and the capitalization can take a couple of tries to get just right.
At least you’re learning more about regexes when you do this. Yes, there’s menial bullshit in coding. There’s menial bullshit in every field. Some of it gets abstracted away (syntax highlighting to help with comprehension), some of it gets kicked around and ultimately does not impress (VB’s drag-and-drop coding), and some of it stays because it’s necessary. Nobody likes doing manual stuff, but sometimes it’s preferable to trying to automate it.
Also, I’ve never heard of anyone paying $20 a month for the privilege of not writing in cursive, or being unable to write because they don’t have internet. Something to think about.
You’re not just making sure you write the letters correctly, you’re also following the syntax rules of the language you’re writing. And while you’re writing, you’re reinforcing those rules in your head.
I get where you’re coming from, but I’ve worked with a lot of bad developers who never got the hang of this even as mid-level developers. On the other hand, I understand the utility of knowing how to do these things for ourselves. There are a number of “black-box” libraries that were just an absolute mystery to me until I tried implementing them myself and began to see these libraries are usually not complex so much as they are thorough in covering edge cases that 90% of users will never care about.
It would definitely be a shame if these tools caused new developers to bypass fundamental skill development. My only hesitation is the number of developers who should’ve developed those skills and never did before AI. There’s something wrong either with how developers are learning or who is getting into development.
I spent a couple weeks trying to use CoPilot and at the end I still had to correct its shitty code, which either hallucinated features I wasn’t implementing, or hallucinated syntax rules I wasn’t using.
We are using CoPilot. As a code-completion engine it is handy. I’m much more skeptical about the new code it writes. Like you, I have not had good experiences with that.
Also, I’ve never heard of anyone paying $20 a month for the privilege of not writing in cursive, or being unable to write because they don’t have internet. Something to think about.
You’re right. Tool access is certainly something to think about. I have more nuanced thoughts, but I don’t want to disagree just to disagree, you know?
On the other hand, I understand the utility of knowing how to do these things for ourselves. There are a number of “black-box” libraries that were just an absolute mystery to me until I tried implementing them myself and began to see these libraries are usually not complex so much as they are thorough in covering edge cases that 90% of users will never care about.
Yeah, that’s one of my big fears. Not necessarily losing my job to an AI, but AI exacerbating existing bad practices.
When I started my current job, we had one rock star coder responsible for a fairly fiddly piece of our product. He went heads-down for two weeks and churned out pages of densely-written python without comments. It did what it was supposed to do, flawlessly. He left the team shortly afterward to work on a bigger project, and we got word from the higher-ups that we had to support a new feature upstream in that code. And then another. And so on. Nothing’s commented. Everything’s over-optimized. We eventually ended up just cross-compiling the upstream logic and using that in our stack because it was easier than using his impenetrable stuff.
In the end, we had to fix it with menial, boring, aggravating manual work anyway. We got ourselves into that situation without AI, but I could see something like that becoming more prevalent. And that was working code. Imagine getting a SEV, and everyone on the blame list shrugs and says “idk, I had CoPilot do it.”
It would definitely be a shame if these tools caused new developers to bypass fundamental skill development. My only hesitation is the number of developers who should’ve developed those skills and never did before AI. There’s something wrong either with how developers are learning or who is getting into development.
Yeah, this is part of it. There’s maybe the science of programming and also, for lack of a better term, the craft: writing maintainable code, handling a SEV, thinking in terms of uptime, setting things up to be reverted easily, shutting down neurotic code reviewers, testing your code… stuff like that. Universities are good at the science part. Internships, theoretically, handle the rest. This isn’t an AI issue, but I could see AI making this problem a lot worse.
I’ve been doing it for more than a decade without help, I’m not any better at spelling or misclicks
And to be clear, I can do it - I just really, really don’t want to. I hate it so much, my eyes glaze over and I have to force myself every second of the way. It’s not interesting, there’s no puzzles involved… It’s basically data entry
It’s not interesting, there’s no puzzles involved… It’s basically data entry
So? Show me an industry that’s 100% interesting all the time. Artists still have to stretch and gesso their canvases. Rock stars still have to deal with band drama and touring logistics. Directors have to work their budgets and wrangle big egos. Why should software, which is basically using fancy math to tell the dumbest guy in the room exactly what to do, be any different?
There’s this awful idea that everything should be fun and nobody should struggle with anything or be forced to do anything menial. We want to be instant experts without going through the boring or hard stuff. And we’re willing to offload more and more of this onto proprietary black boxes in exchange for… what?
So I should suffer just to suffer? You listed a whole lot of things that they hire people to do just about as soon as they can so they can. And offloading that let’s them do their actual job better
I work with black boxes all the time. When I have a black box, I poke and prod it until I understand how to make it do what I want. And this particular black box was interesting, so I decided to open it up and learn how it works
That’s the essence of software development. My job is not typing or data entry, my job is to trick a rock into doing things humans don’t want to do
Lol
AI writing code for me made me the software architect I always dreamed of becoming.
I fucking LOVE to think about a hard problem for days, planning, researching, comming up with elegant solutions, doing quick POC, thinking what needs to be refactored for it to scale to a real life scenario, then documenting it all in a way that is properly communicating the important aspects in an easy to understand way. It’s so exciting!
And I fucking HATE having to sit down and actually type out the solved code for hours and hours. It’s so boring.
Best 20$ per month subscribtion I’ve ever had.
Yep. It’s gonna be $20 forever, too. Have fun!
Lol. Lmao even
It does save a lot of time and effort, and does lead to better code in the hands of a skilled developer. Writing out thorough test code and actually doing proper test driven development suddenly becomes a lot less onerous.
Their graph also has no numbers and is just there to help visualize the difference they’re referring to.
To the first part, I agree. A skilled developer who can quickly separate the wheat from the chaff can get a boost out of AI. I’d put it at around 5-10%, but I’ve had some tiny projects where it was 400% boost. I think it’s a small net gain.
As for your second point I just have to disagree. There are no numbers but it is clearly selling the idea of the majority of code being AI generated, and that’s bullshit whether it’s an outright lie with numbers, or merely vaguely misleading. It’s like when someone cuts off the bottom of a graph to make relative change look huge. It wants people to glance at it, get the wrong idea, and move off without curiosity.
To the first part, I agree. A skilled developer who can quickly separate the wheat from the chaff can get a boost out of AI.
It takes less time to just write code than to babysit an artificial dumbass.
AI-assisted coding […] means more ambitious, higher-quality products
I’m skeptical. From my own (limited) experience, my use-cases and projects, and the risks of using code that may include hallucinations.
there are roughly 29 million software developers worldwide serving over 5.4 billion internet users. That’s one developer for every 186 users,
That’s an interesting way to look at it, and that would be a far better relation than I would have expected. Not every software developer serves internet users though.
I’m 90% sure it’s something to do with the stock market, buy backs and companies having to do cryptic shit to keep up with a fake value to their shares
What do you expect? Half of these decision makers are complete idiots that are just good at making money and think that that means they are smarter than anyone who makes less than them. They then see some new hyped up tech, they chat with ChatGPT and they are dump enough to be floored by it’s “intelligence” and now they think it can replace workers but since it’s still early, they assume that it will quickly surpass the workers. So in their mind, firing ten programmers and saving like two million a year, while only spending maybe a few tens of thousands a year on AI will be a crazy success that will show how smart they are. And as time goes on and the AI gets better, they will save even more money. So why spend more money to help the programmers improve, when you can just fire them and spend a fraction of it on AI?
Also is substack the new meduim? I cant keep up with these freemium wordpress/blog clones.
Its not that dumb as you think, its way dumber.
