Vibe-coded tools vs. the law of comparative advantage
Just because you can rebuild SaaS with AI doesn't mean the economics say you should.
Vibe-coded solutions replacing SaaS ignores the law of comparative advantage.
Coding has been one of the strongest applications of AI so far. Code is fundamentally verifiable — it either compiles or it does not — and there is an enormous amount of training data available.
With the rise of AI agents generating dashboards, and tools like Claude Cowork becoming mainstream, a common narrative has emerged that SaaS is on the chopping block.
Yesterday I tried to spin up a small interactive playground to help me practice AVL tree rotations visually. I could not find anything that matched what I wanted, so I tried using several vibe-coding platforms to generate a quick prototype.
It was rough.
I tried multiple prompts and specifications. After about an hour and a half I was essentially back where I started. Had I just done it by hand, I would have been better off.
At that point I remembered something Andrej Karpathy said on Dwarkesh’s podcast: LLMs struggle with novel or less common problems.
An interactive AVL playground is not groundbreaking, but it is not a common production pattern either.
CRM systems are a good example of where vibe coding works well. There are endless open-source CRMs and huge amounts of training data. If you ask an AI to generate a CRM dashboard, it will often produce something that looks surprisingly sophisticated.
But that does not mean the underlying system is robust.
It also ignores something much more practical. Time.
If your productive time is worth $1,000 per day, your hourly rate is $125.
Is it worth spending 20 hours vibe-coding your own CRM instead of paying HubSpot or Pipedrive?
Probably not. Just like you can grow your own tomatoes and raise your own cow, but you choose to buy the ingredients from the supermarket because it’s cheaper.
Comparative advantage still applies. Just because something can be built does not mean it should be.
AI will absolutely make building software easier. But the economics of specialization still matter.
Sometimes the highest-leverage move is still to pay for the tool.