And the Surprising Potential of Having Computers Write Their Own Code
Not too long ago, I started on a journey with “vibe coding” when I was innocently asking my AI assistant Claude if it would be feasible to write an in-ear AI assistant than ran on local hardware and emulated Jarvis from Iron Man (the movie starring Robert Downey Junior). It was an idea for an AI assistant that I had years ago before the advent of LLMs that had been acutely interested in, but totally unable to implement. Now I wondered if I could accomplish it.
I asked Claude, innocently:
How much work would it take to create a “Jarvis” in your ear earpiece that connects to a home server and acts as a web search assistant using voice commands?
Claude gamely replied that the effort would be substantial but very doable, then asked:
Want me to sketch out actual starter code you could run this weekend?
And so it began. I’m now on version 1.1 (over 60 revisions later) of the project and though it’s morphed a little bit from its aspirational beginnings, it’s still some amazing software, written entirely by Claude. It runs a local LLM from Google running through Ollama on my laptop to which I can give typed text commands, or more fun – voice commands through a bluetooth headset. It’s actually pretty smart too.
Usually I let Claude do the programming because it actually understands the code, and even if I did understand the code, Claude is so much faster.
Might I add I haven’t paid for a Pro subscription this whole time? I use the artificial limits of a free subscription to help me maintain a coding/life balance.
This makes me wonder: Is this the future of computer programming? Where we delegate coding and technical skills in the digital world to AI?
Starting a Project vs Finishing it
I recently happened on a video by Sara Dietschy (Rhymes with Peachy) on Youtube where she details the vibe coding process she used to arrive at a finished project with the help of Claude Code running in Cursor.
She originally started building her project using v0.app but switched to Claude Code when her app broke halfway through the process because whatever AI model she had been using was not up to the process of debugging her app (Probably an Opus 4.8 model actually).
But when she finally involved Claude Code it was able to immediately spot her problem for her which had actually been simply missing API keys which had been corrupted.
Now Sara is not a coder: In fact she is a CS dropout who has – according to herself – forgotten most of what she knew about programming. That being said, she has some technical ability in my estimation.
This technical ability is what makes or breaks the difference between being able to start a project and being able to finish it with modern tools right now. And Sara found this out first hand when she didn’t simply give up and instead switched to a tool she knew was more powerful.
Where This is Heading
But I believe the time is coming when semi-technical people like Sara will be able to create full products and software projects without ever touching any code at all, simply by leveraging the use of AI tools like Claude Code or V0.
Now in my case with my Jarvis project, I found that I was actually able to understand most of what Claude was telling me because I have a coding background, not all users are going to have this context. This makes it a lot easier to guide the model doing the actual coding.
Eventually however, AI apps will be able to handle the whole process on their own without human intervention even. And it’s not that far away.
Is this something to welcome, or something to worry about?
The real question is: Is this a good thing, or something that we want to avert? I’m currently pursuing a career in software development and was initially alarmed not too long ago that super-capable AI models would push me out of my chosen career.
But I’m not so worried about that anymore as I can see with my new experience with “vibe coding” that there will always be a place for a human in the loop. And even for applications where humans get pushed out of the loop. There’s always something else to put your focus on. After all, humans created AI in the first place.
In fact, I have found the simple act of telling Claude what to do to be refreshing in and of itself. Seeing a program or project come to life in a very short time can be a genuinely compelling reason for letting the AI agent take the wheel sometimes. But that’s my opinion.
Fear is the mind-killer to quote Frank Herbert. You really have to trust God at some point to escape existential dread. But I digress.
I recently wrote a sci-fi story titled Where the Light Shines in which humans live in a dystopian future where highly intelligent machines run everything and exist in sealed off rooms where they fully maintain themselves in every way (though they’re simply referred to as the daemons by the vulgate).
In this story the machines are described as somewhat under-equipped to fully take over human tasks and do a good job.
“Turns out super-intelligent computers couldn’t fix what happens when systems grow too large for anyone to be responsible for them anymore.
But that was largely narrative worldbuilding at work – I didn’t really have an opinion one way or another at the time.
Takeaway
If automation has mostly displaced the work we didn’t want to do anyway, and often done it better — what exactly are we afraid AI is going to take next, and is it the technology we should be worried about, or our own fear of it?
In my opinion, fear is usually the enemy, not the savior.
Note: If you’re interested in checking out the app I built, you can see the code for the script on my github page, under the new name Friday.
Timothy Grindall is a technology writer and self-taught computer programmer, photographer, filmmaker, thinker, and writer who wants to see the world change for the better.
Leave a Reply