Is the cost really approaching zero though? It still needs a skilled dev to supervise. And any speed is offset by the cost of tokens. Cheaper yes. But approaching zero? Not sure.
I don't think it could literally ever go _to_ zero, but think about it like the example in the post. All these businesses built around the cost of shipping before the shipping container was invented. One of the inputs to their business fell by a factor of 30, and if shipping was a major line item, it would definitely feel like costs fell to near zero.
Now think about how big of an expense software development is for most businesses. By what factor does that fall once more people are comfortable with Claude Code/Codex/etc?
And on needing skilled devs to supervise, think about how many businesses are built on top of the macros in that one spreadsheet that one person did a decade ago. Building useful things at a higher quality bar than that for your business with Claude Code is already available today for basically everyone at $100-$200/month.
I think there are two factors that could keep software costs up. Quality and how fast technology changes. AI software quality is still questionable, unless someone who knows what good software looks like is doing it. Does the random MBA with a $100 month subscription know to ask for automated tests or know to ask the AI to hash passwords? When these things go wrong cost goes up in order to pay someone to fix it.
The other thing is technology development. You mention ten year old macros, and that may work in excel but software ages far faster. Yes, we can rely on skills or mcp servers, but those have dangerous risks if not properly vetted and could only be relevant for a set period of time.
All these things being said, I still agree with you. I think AI is opening up markets that would not be previously viable due to their small size. I just think, that since we are now 3 years into being 6 months away from software devs being irrelevant, that it's still going to take a trained developer to build software that works and works well.
Is the cost really approaching zero though? It still needs a skilled dev to supervise. And any speed is offset by the cost of tokens. Cheaper yes. But approaching zero? Not sure.
I don't think it could literally ever go _to_ zero, but think about it like the example in the post. All these businesses built around the cost of shipping before the shipping container was invented. One of the inputs to their business fell by a factor of 30, and if shipping was a major line item, it would definitely feel like costs fell to near zero.
Now think about how big of an expense software development is for most businesses. By what factor does that fall once more people are comfortable with Claude Code/Codex/etc?
And on needing skilled devs to supervise, think about how many businesses are built on top of the macros in that one spreadsheet that one person did a decade ago. Building useful things at a higher quality bar than that for your business with Claude Code is already available today for basically everyone at $100-$200/month.
I think there are two factors that could keep software costs up. Quality and how fast technology changes. AI software quality is still questionable, unless someone who knows what good software looks like is doing it. Does the random MBA with a $100 month subscription know to ask for automated tests or know to ask the AI to hash passwords? When these things go wrong cost goes up in order to pay someone to fix it.
The other thing is technology development. You mention ten year old macros, and that may work in excel but software ages far faster. Yes, we can rely on skills or mcp servers, but those have dangerous risks if not properly vetted and could only be relevant for a set period of time.
All these things being said, I still agree with you. I think AI is opening up markets that would not be previously viable due to their small size. I just think, that since we are now 3 years into being 6 months away from software devs being irrelevant, that it's still going to take a trained developer to build software that works and works well.
This feels a lot like convergence in an article about divergence when I consider the recent writings of https://www.substack.com/@limitededitionjonathan and https://www.substack.com/@russmiles
All in all, very interesting and a very apt example!
Haha :) I'm not familiar with them, I'll have to check out what they're doing. Thanks for sharing!