One of the things I get asked most right now is this: will AI replace software developers?
Or the variations: how many developers are we really going to need with AI, or when will AI be capable of doing all the work?
If you’ve recently tried vibe coding, or using a ready AI interface to generate webapps or projects without actually writing any code, you may think such a day is soon.
After all, as Andrej Karpathy (a founding former employee of OpenAI and the person credited with coining the name) posted in February on X, you can develop some apps right now without touching a single line of code.
He wrote:
“… I just talk to Composer with SuperWhisper so I barely even touch the keyboard. I ask for the dumbest things like ‘decrease the padding on the sidebar by half’ because I’m too lazy to find it. I ‘Accept All’ always, I don’t read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it…”
Karpathy is himself a highly acclaimed computer scientist, so this is for his own amusement, largely. Throwaway weekend projects, as he calls it.
But still, he writes: “I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”
In today’s newsletter, I want to consider these things together, or vibe coding (what it is, what it can do) and the extent to which this sudden and sometimes impressive capacity reflects on a fast-changing reality for software developers.
And I’ll get back to answering those leading questions, from my point of view, too.
What’s Behind Vibe Coding Trends
Maybe you’ve already heard of vibe coding, but if you haven’t, the best introduction is giving it a try.
Some of the popular tools at the time of this writing include Cursor, Bolt, Claude, Lovable, and Replit, and you can try most for free.
Or you can also take a look at any number of videos, like this one by veteran developer Jamie Marsland. He showcases a truly impressive website interface which he created in just two hours. The entire front-end he made with Lovable.
Here, he demonstrates making a change (the image that appears when hovering, as an example), and the AI system provides a preview, pushes it to his live site, and even updates his GitHub repository. All in seconds.
Marsland is himself a software developer, but you can also read about non-developers doing things like this, such as in a great piece by Kevin Roose for The New York Times (see below).
(You may remember him for gaining national attention two years ago when a chatbot declared that it loved him, wanted to be alive, and wanted him to leave his spouse.)
Roose, who writes that he “can’t write a single line” of code, uses Bolt to build an app that takes a provided photo of the contents of his fridge and provides suggestions for making his son’s lunch.
And in ten minutes, he had it working. (You can try that, too, here.)
Taken together, it’s easy to understand the hype.
No-Code AI Development: Where It Impresses
These demonstrations often look too good to be true. As Marsland demonstrates, he can simply prompt the AI to ask for more accessibility, and it makes improvements to the site directly for him.
In Karpathy’s case, he uses a voice interface, so he doesn’t even need to type, let alone write the code.
Need a rapid prototype? Vibe coding tools can throw one together for you far easier than I could have imagined possible, ten years ago (especially if you have a ready example to point to).
Non-coders can now create examples of what they have in mind without the need for additional technical help, which I can imagine greatly aids communication between teams, as well as potentially increasing awareness.
Want to get your feet wet with development? AI-powered app builders work with your concepts from just ideas, allowing you to see what is needed to bring it to life.
From a learning point of view, this technology has incredible potential. But it won’t necessarily give you a best-case example to work from.
Shortcomings and Limitations in No-Code AI Tools
Now for the reality check. Just as a UI may be exceptional to explore but may fail to deliver core functionality, vibe coding solutions have their own abundant limitations at present.
In his article, Kevin Roose provides several examples where he has been unable to vibe code a functional solution in areas you may think would be easy for AI (such as responding to emails in his style).
I don’t need to cover hallucination or mistakes except to point out, again, that it makes them. (And yes, they are improving.)
More importantly, as with image and video generators, AI vibe coding tools excel at providing variations on things that already exist. Combinations and customizations in other words.
A more critical issue is security. Many cybersecurity experts are greatly unnerved by the rise in no-code services. They warn that they can expose organizations to serious risk if used for projects that aim for real-world application or integration with existing solutions.
To be meaningful, applications need data, and any AI code that comes into contact with user data still must be carefully checked. And debugging wholly AI-generate code can be difficult, let alone when compounded over time.
In the best cases, vibe coding projects thrill with their ease and dazzle with their apparent polish. But they are not finished products, even if they often look like them.
No-Code vs Traditional Coding: Developer Impact Outlook
So what does this mean for developers?
Let’s go back to the positives: AI coding solutions mean real efficiency improvements. Research abounds showing that AI copilots for application development are now giving a 10–30% boost to productivity. I’ve heard developers say their pure coding time is cut in half in many cases.
Paired with good standards and processes, this can mean improved code quality, too. Overall, about an hour a day, on average, can be saved with AI assistance, and this tracks with Copilot estimates that put the benefit to developers at 4.5 hours a week at present time.
It’s no wonder that investments in this arena neared $1.6 billion in 2024 alone, triple the amount from 2023, according to PitchBook.
AI change has come for developers, and it’s come faster than it has for most other professions.
But shaving off an hour or even two a day, significant as it is, is a far cry from non-developers building business apps on their own.
As Karpathy wrote, AI can get something on its feet by itself, given enough iterations, but that’s for weekend play at this point.
The very benefit of AI coding solutions—generating functional code, fast—is also one of the reasons human coders are still so necessary. Because while it may sound good, and even look good, it isn’t necessarily right, or safe.
And the more that’s there, and the deeper it goes, the harder it is to verify and secure.
Commonly we hear that, with AI agent–driven development coming online now, these tools will soon be writing 80%+ of all new code, functioning like an insanely fast, and extremely talented (though inexperienced), intern. But that final 20%, not to mention the proofing of the rest, is utterly critical.
This means learning new skills, and that’s something developers aren’t shying away from. In a recently released survey of software engineers conducted by Wired, three in four indicate they’ve used AI, with the vast majority using it at least weekly.
While those surveyed remain concerned that leadership has overly idealistic views on AI and what it can do (inspired, no doubt, by the highs of vibe coding projects like the ones detailed above), veteran coders remain the most likely to fully integrate AI into their workflow, going beyond browser interfaces.
They’re also more evenly split in their outlook.
Now back to my leading question.
Can AI Replace Software Developers?
Yes, of course it can.
But you don’t need me to tell you that.
Developers translate human asks into machine actions, so yes, of course it is theoretically possible that AI systems will be able to do that.
Theoretically.
But this is not the question that matters. The real question is will it?
And the answer there is no.
Not now. Not anytime soon. This according to experienced pros, business leaders, AI developers, and my own experience to date.
There are other, more significant concerns for developers than replacement, with the process being, as Steve Lohr describes it, “more likely evolution than extinction.”
First and foremost, it’s about learning. And not being lulled to sleep by the apparent ease. (Which may be harder than it sounds.)
One intriguing question to me is what schools do with this evolution. In decades past, some programs avoided languages even as low-level as Java, not allowing even popular libraries to be used at first.
The argument is “learn to walk before you run.” Or better, crawl first. Then walk.
In this mindset, it’s easier to move from C to Java, than from Python to C++. (Though of course, many fine coders have done it both ways.)
Does this dynamic change with AI? Or does the same thinking still apply? Using AI well is itself a highly necessary skill, but it may be helping more experienced developers most in terms of quality, if not in terms of sheer speed.
Overall, AI is a great leveler for those with less experience, as we discussed in this article. But the irony is that it doesn’t replace the need for those very experts to check the work that’s being done.
This begs the question: will there still be sufficient expertise among the developer workforce with fewer positions available for junior developers, AI acting as interns, and no-code solutions coming online to take the lightweight tasks?
PTP Represents Quality Developers
I’ve been working with software developers for more than 27 years, and it’s been my pleasure to represent many of the hardworking, talented people who make our systems not only go, but also efficient and safe.
If your organization is in need of such people, contact us.
While AI automation is here, powerful, and in some arenas truly transformative, it is not going to take away the need for this experienced human capacity.
As I mentioned above, AI solutions do best at creating things that already exist, or by creating combinations or customizations thereof.
To open new doors, innovate, and create truly new things, we still need talented people. And that’s not going to change anytime soon.
References
How Software Engineers Actually Use AI, Wired
Not a Coder? With A.I., Just Having an Idea Can Be Enough. and A.I. Is Prompting an Evolution, Not Extinction, for Coders, The New York Times
10 professional developers on vibe coding’s true promise and peril, ZDNet
