For those of my readers not familiar with the term "vibe coding": it is a common parlance term describing the latest evolution of AI assisted programming. First, AI acted like a fancy autocomplete. Next, AI wrote larger chunks of code and provided insights, but was largely limited to a single file. With vibe coding, AI now works across an entire codebase, understands project structure, and coding responsibility is fully handed over to the AI (with the human guiding it). Vibe coding is the next phase in a step-by-step growth of AI responsibility in the coding process.

Where we are (subject to being outdated prior to writing)

We are in the beginning stage of the software industry being completely upended.

You may have seen in my essays a combination of hope and enthusiasm for AI mixed with (what I hope comes off as) pragmatic pessimism. This essay is going to lean heavily toward discussing potential.

I had a discussion not too long ago with Harper Reed that was eye-opening. (He's cool and I owe him for his time and insight - go check him out.) I brought my more pessimistic side into our discussion. He challenged me in an important way, saying in effect: if the AI isn't able to accomplish what you want, it's more likely to be a problem with your prompting than with the AI itself.

He walked me through his coding process (which he's kindly detailed publicly on his blog). Though I was still skeptical, I could see he was employing techniques that were new to me. The process seemed feasible and I was open to the possibility it would work. So I was going to have to try it out.

I did, and it worked. And it was amazing, surreal, scary, and wonderful. And magic. All at the same time. It reminded me of the story about accountants reacting to the first PC-based spreadsheet program - "he started shaking - he said, 'This is what I do all day!'"

It's also...quiet.

Old-style human coding for me is like balancing a tower of Jenga bricks inside my brain - layer upon layer forming a stack of thoughts requiring a light and careful touch. And the bricks are made of lead, practically eager to crash down and bring entropy back to my thought processes.

Vibe coding is peaceful, even elegant. (When it works.) No careful attention balancing is required, you just need time and focus on your end goal.

It's of course not perfect. It's like talking to a brilliant scientist who can't tie their shoelaces. Moments of insane hyper-productivity intermixed with moments of utter head-smacking cries of WTF. The aigents aren't taking over, yet. But they've come a long way in a short time.

I was happy to discover that vibe coding prompt techniques are useful for much more than just programming. I recently completed a non-programming task which normally would have taken a couple analysts a couple weeks to generate their deliverable. With the help of an LLM and careful prompting, I went from idea to ready-to-send report in about 3 hours. Words cannot express my amazement at the productivity.

What this means

If that story makes you fear for the future of employment and AI taking jobs, I have both good news and bad news.

The good news is, the task I referred to was an augmentive task, one that either got done cheaply or didn't get done at all. I suspect many highly productive AI tasks are like this. Without AI I would have made a gut-level decision and communicated it verbally in an attempt to "sell" my thoughts. Using AI, I corroborated my intuition with objective data and presented a deep thought in a consumable way. (In other words: No analysts were harmed during the production of this essay.)

The bad news is, yeah, there's some pretty amazing productivity gains. Quoting Neal Stephenson quoting Marshall McLuhan, "every augmentation is also an amputation". The incentive structure of modern business - whether or not you like or agree with it - is to minimize human cost to every extent possible. And so there are going to be lost jobs, stress, and heartache. And personally I worry that we don't have skillsets ingrained as a society to deal with this turbulence in a humane and positive-sum way.

The zero-sum approach is: our people are more productive so we need less people. The positive-sum approach is: we need AI to help make our people indispensable. Sadly not every business will approach this in a positive-sum way.

My biggest hope is that with the workforce reductions we'll see greater creativity - new startups, new ideas, and new lines of business - that will replace and augment what's lost. But in truth, I don't know what happens to the workforce from here.

The rise of super-humans

The term "super human" is inherently hyperbolic but it's really the best way to describe it - the 100x productivity hack. I think about this in terms of software engineering but it may apply to other professions as well - ping me if you have good examples.

The vibe coding process has the potential to yield (intuitively, based on my personal experience) around a 10x productivity gain in software coding, making it a remarkable violation of Fred Brook's no-silver-bullet axiom.

But, as I've said before (link omitted because the rest of the essay aged like milk 😬) engineers don't truly spend all that much time coding. So while a 10x reduction in coding time sounds great, it doesn't translate into 1/10th the engineering cost because you still need to staff the other engineering functions, such as product management, SME/analyst, project management, graphic design, QA, etc. And then your engineer has spend time communicating with all those people. And humans communicating with each other, as fun as it may be for some, is time consuming. It turns out that transmitting thoughts between human brains is hugely inefficient.

But, what if you have one of those rare breed of people who's able to work across all those domains in a single brain? One person working solo who can research the customer, define the requirements, design the interface, design the software, code it, verify it, and ship it? Able to work through all of that communication inside their head without having to write it down or spend time talking? I can't say exactly how much time that would save, but I think 10x is a reasonable guess.

Now, there are very few people who are innately able to do all those things well on their own. But with AI assistance that venn diagram grows a fair amount larger. And that is how we achieve the theorized 100x super-engineer who can bring down the entire engineering cost by 10x.

Such people are out there. We just have to learn how to identify them - distinguish them from those paltry obsolete 1x-ers. And then set them up to succeed.

While they do their amazing thing, the rest of us will figure out how to keep ourselves busy.