I’m trying to understand this one and it doesn’t really make sense to me. Or maybe —-
What is software? How do you delete software and start over. I think we’re differing on our definitions of what software is.
If you have a clear spec, then maybe we have like, immutable software, like —- you don’t upgrade dependencies, because you can recreate software from spec. But then to me, software is the spec not the code. Like, do you just etch a sketch your product every year going back to customers to figure out how to rebuild from scratch? I don’t think that’s what’s being suggested.
I suspect what’s being suggested, is that if you are able to codify all the business logic, edge cases, and optimizations, then you can generate software from that. But to me that’s what software is, not the code.
But if you don’t have a clear spec? Isn’t one of the reasons you don’t rewrite legacy mainframe software because the spec is unclear? You don’t want software that recreates that old software, you want software that makes new assumptions.
Also - I guess - it seems like we’re assuming that Claude writes perfectly optimized code, is that right?
But then, why 1 year shelf-life? Why are you keeping software around for a year? Why not regenerate for every deploy? One year seems like this weird middle ground.
Take I dunno, the OpenSSL library, are you having researchers re-imagining the encryption algorithms every year?
There’s something about the way people talk about new paradigms that seems in bananas opposition to software engineering best practices of the past ten years. But like BANANAS opposition.
And I feel like people think I’m being a negative Nancy, but nothing in this doc feels like a realistic path to the 100x everyone wants, and I want the 100x gains! Do people forget that when the rubber hits the road this software has to run in a hostile environment?
The solution in this doc seems to be; there will be 100 software companies run by the 100 180 IQ engineers in the world. Do we not care about the bus factor?