The future of programming: why English is becoming the new programming language
Jensen Huang, CEO of NVIDIA, recently made a statement that caught the attention of the tech world: “English is the hottest new programming language.”
Is Huang right? I think so. Not only because 30 to 40 percent of all new code on GitHub is now written using AI. The story is more complex than that.
How AI is fundamentally changing the role of developers
After about 30 days of programming with AI, during which I managed to build two cool products, I can say one thing with certainty: programming is taking on a whole new meaning.
Research shows that programming teams that work with AI are approximately 26 percent more productive and, above all, develop more balanced solutions. There are also studies that claim the opposite: developers who used AI coding tools were on average 19 percent slower. But in all honesty, all the studies I read were outdated before they were published.
The MCP server concept has been around for less than a year, and Claude Code has only been widely available for three months. It was precisely this combination that showed me the true superpower of AI for software development.
I expect AI-augmented teams to be able to go from idea to product five to ten times faster, provided they are set up for this. This requires something different from developers: conceptual thinking will become much more relevant than actually typing code.
From technical complexity to business insight
Even more interesting is how input from the rest of the organization will be processed. For non-technical people, it is often difficult to estimate how difficult it is to build new features.
But if I give AI access to our codebase and explain what business functionality I want, the AI can make a pretty good estimate of which systems it will affect and how complex it will be. Even better, I can ask how I can get a similar feature with far fewer development hours.
A marketing team can already have a prototype built and tested with users based on market research, without a single line of code being written by a developer. The feedback loop becomes shorter and faster.
However, programming with AI also has a downside.
The downside: why AI code can be dangerous
AI loves writing new code. This is mainly because, when it comes to large, complex codebases, AI simply lacks the architectural awareness needed to build maintainable systems. The result? Studies show that AI-edited projects contain up to eight times as many duplicate code blocks.
The cost of new AI code
That generated code also has a cost. AI data centers are expected to consume around 6.7 to 12 percent of all US electricity by 2028. And 80 to 90 percent of that generated code doesn’t even make it to the production phase. Ouch.
On the other hand, there is a striking paradox: for years, code efficiency was hardly a priority. Servers — or “hardware,” in technical terms — were cheaper than development hours. The result? More users meant more hardware.
AI is changing that. Efficiency is becoming achievable again. You can request energy-efficient algorithms or let AI work in languages that you don’t master yourself, but that are perfectly suited to your project. The result is faster, more efficient code.
The hallucination problem: non-existent software and security risks
Even more worrying are the hallucinations. Thirty percent of AI-generated code contains at least one “hallucinated” package dependency, i.e., software libraries that simply do not exist. Of the 576,000 code samples examined, 440,000 contained references to software packages that cannot be found anywhere.
This opens the door to package confusion attacks. Cybercriminals can create these non-existent packages and fill them with malware. Developers who blindly trust the AI code will then literally install malware.
Approximately 40 percent of all AI-generated code contains security vulnerabilities. The code looks fine and usually functions properly, but it contains security holes that are only discovered later — often by hackers.
Security can quickly become a bit of a red herring, because code written by humans also contains many errors. If you regularly incorporate a thorough clean-up round into your AI workflow – making the code neat and readable – and analyze the code as it is being written, you can eliminate a lot of security errors from the systems. So be aware of the security risks and adjust your workflow accordingly.
Code review in the AI era: hybrid and essential
This makes code review — where a (senior) developer reviews the code of a colleague (junior) developer — more important than ever, but in a completely different way. Traditional code review is being replaced by a hybrid process.
AI tools can scan code for syntax errors, style issues, and known security patterns in seconds. But for context, architectural decisions, and subtle security risks, we still need humans. It’s as if AI does the spell check, but humans still have to assess whether the story structure is correct.
We need AI to check AI. Trust, but verify.
Legacy code modernization: where AI really excels
Where AI does excel, you see serious results. Modernizing legacy code is the best example of this.
For the average financial institution, around 70 percent of IT capacity is spent on maintaining legacy systems. Systems written in COBOL, sometimes decades old, still form the heart of banks worldwide.
AI is exceptionally good at rewriting old code into modern architectures. AI understands the logic of old systems and can translate it into something new. This means you can use AI to pay off your technical debt, freeing up more time for innovation.
The silent revolution in business processes
This innovation is not limited to development. Workflow tools such as N8N, combined with small AI-generated code snippets, make it possible to automate many administrative processes.
Where a human colleague might be able to handle 20 customer service tickets per day during business hours, a well-trained AI can analyze hundreds of tickets 24/7. Accounting processes, VAT returns, HR leave requests: there is a lot that can be automated with AI workflows.
However, optimizing such a workflow is still very complex at the moment and reserved for people with a technical background. So while there is great promise, I think it will be years before many of these administrative processes are truly automated.
Who actually benefits from the AI revolution?
But who actually benefits from the AI revolution? That’s an interesting question. Because there are still a lot of people working on those administrative processes. At the moment, it’s the highly educated, adaptive knowledge workers who are benefiting from the AI revolution. Think of the marketer who can now build prototypes and the consultant who can produce reports much faster.
For administrative colleagues, junior developers, or customer service experts, however, it mainly feels like a threat. Companies must therefore invest in their people: give junior developers room to grow to mid-level, and teach administrative colleagues how to use AI. That way, they too can keep a job that suits them.
What we lose: the human factor
And this is where this story touches on something that really concerns me. When we talk about AI, we mainly talk about faster, more efficient, and cheaper. But I rarely hear the word “more human.”
The person on the other end of your chat or phone call will increasingly be an AI and less and less often a human being. If the human aspect disappears from our interactions and from our work, what will that do to us as a society?
The generation that never learned to search
A generation is growing up that has never learned to Google effectively, because they can ask ChatGPT questions directly. That has never debugged code, because AI does it right away. That has never learned to structure their thoughts, because AI rewrites their texts.
Is that a bad thing? Yes! Initial research suggests that it is extremely detrimental to our brain development. One could argue that the automobile also replaced the horse, but that is a matter of transportation. AI is replacing our thought process. These are two entirely different matters.
The choice we’re facing
So where do we end up? With a world in which AI does all the routine tasks and people can focus on creativity, strategy, and genuine human connection? Or with a world in which we have become so efficient that we have forgotten what we were actually living for?
The technology is there. The choice is ours. And we don’t make that choice in boardrooms or at tech conferences. We make it every day, with every decision about how we use AI.
Do we choose AI that gives us more time for each other? Or AI that makes us even more productive in a race that no one can actually win?
I don’t know the answer. But I do know that we need to think about it carefully. Because the tools we build and use today will determine what the world looks like tomorrow.
Blog series on programming with AI
This is the last article in a four-part blog series on programming with AI. You can read the first three articles via the links below:
More stories to read
On our blog we post about a lot of stuff, just go for it and read some posts for your own fun.
Go to the blog
