Years ago I when I wrote software for a living, I had an argument with a colleague, and I tried to explain to him:
The “supported” closed-source library he wanted to use was pretty popular because it was marketed by a huge company with a marketing department, or because it had a first-mover advantage, or because there were training events and books built around it, etc.
The unsupported free open-source library I wanted to use was the most popular library of its kind in the whole world. And it got to that position without any of those advantages.
What does that suggest about their relative usefulness? The world of open source is closer to being a real meritocracy. The number one app or library is probably number one for non-structural reasons.
Eh, Claude’s cutting edge frontier models are definitely better than the good open source models which lag a bit behind. The good open source models are still useful though but you’d get noticeably better performance with the closed model which is why even companies that are perfectly capable of locally hosting an open model choose to pay anthropic a premium.
Maybe but that wasn’t my point. My point is that a lot of people now invest a LOT of resources, being token, money, time, etc to invent the wheel again. Instead of relying on e.g. Drupal they’ll “generate” yet another CMS which will work (for a while, in theory) not because it’s a good idea (IMHO it’s not) but because it’s been marketed as doable and even “better” on some aspects (e.g. customizable).
Ah, you’re referring to local rewrites of utilities that already exist?
I agree that agents are making more in house utilities which can be wasteful. The shift certainly isn’t helped by the increase in supply chain attacks though.
I think the main reason most companies choose closed source is because management gets a hard-on for the thought of having someone to complain to. If they can’t call meetings with someone responsible and demand a quick fix, what use do they still have? All you can with open source is fix it yourself or create an issue. Neither requires a manager.
The point is that it is impossible to have support of every single software you use in-house. So it is better to outsource it to companies who have specialized support on hand 24/7, and who have been solving those kinds of issues every single day of the year. They don’t need to flip through the documentation in order to solve it.
In companies, a problem that causes the entire company from being unable to generate profit for 24hours costs way more than a support contract.
This support stuff (for problems that occur often enough for anyone to build experience) is one thing I’ve found AI is pretty good at. Even more obscure issues, and especially open source stuff because it was probably trained on the source code as well as any public support forums.
So this might stop being a factor, or as big of one, when an intern and an AI can figure out most issues in an afternoon.
Correct. Also, they need someone to delegate the responsibility to. They are mainly concerned with not being held responsible for any potential fuck-ups. If they can say “the vendor did it” they can deflect the blame. Unfortunately that’s how making a career in the corporate world works for the vast majority of people. You advance by avoiding getting blamed for mistakes, not by brilliance or competence.
Years ago I when I wrote software for a living, I had an argument with a colleague, and I tried to explain to him:
The “supported” closed-source library he wanted to use was pretty popular because it was marketed by a huge company with a marketing department, or because it had a first-mover advantage, or because there were training events and books built around it, etc.
The unsupported free open-source library I wanted to use was the most popular library of its kind in the whole world. And it got to that position without any of those advantages.
What does that suggest about their relative usefulness? The world of open source is closer to being a real meritocracy. The number one app or library is probably number one for non-structural reasons.
And that’s why GenAI for code is gaining popularity.
It’s not because it’s better than free open-source libraries. It’s because it’s better marketed.
Eh, Claude’s cutting edge frontier models are definitely better than the good open source models which lag a bit behind. The good open source models are still useful though but you’d get noticeably better performance with the closed model which is why even companies that are perfectly capable of locally hosting an open model choose to pay anthropic a premium.
Maybe but that wasn’t my point. My point is that a lot of people now invest a LOT of resources, being token, money, time, etc to invent the wheel again. Instead of relying on e.g. Drupal they’ll “generate” yet another CMS which will work (for a while, in theory) not because it’s a good idea (IMHO it’s not) but because it’s been marketed as doable and even “better” on some aspects (e.g. customizable).
Ah, you’re referring to local rewrites of utilities that already exist?
I agree that agents are making more in house utilities which can be wasteful. The shift certainly isn’t helped by the increase in supply chain attacks though.
I think the main reason most companies choose closed source is because management gets a hard-on for the thought of having someone to complain to. If they can’t call meetings with someone responsible and demand a quick fix, what use do they still have? All you can with open source is fix it yourself or create an issue. Neither requires a manager.
What? No!
The point is that it is impossible to have support of every single software you use in-house. So it is better to outsource it to companies who have specialized support on hand 24/7, and who have been solving those kinds of issues every single day of the year. They don’t need to flip through the documentation in order to solve it.
In companies, a problem that causes the entire company from being unable to generate profit for 24hours costs way more than a support contract.
This support stuff (for problems that occur often enough for anyone to build experience) is one thing I’ve found AI is pretty good at. Even more obscure issues, and especially open source stuff because it was probably trained on the source code as well as any public support forums.
So this might stop being a factor, or as big of one, when an intern and an AI can figure out most issues in an afternoon.
That’s exactly my point?
Correct. Also, they need someone to delegate the responsibility to. They are mainly concerned with not being held responsible for any potential fuck-ups. If they can say “the vendor did it” they can deflect the blame. Unfortunately that’s how making a career in the corporate world works for the vast majority of people. You advance by avoiding getting blamed for mistakes, not by brilliance or competence.
Many open source have paid support
And then after they demanded a quick fix it will be swiftly delivered in next decade
A main issue, according to my non-software related work-life experience is also: liability reasons.
Being able to legally blame someone else when shit goes wrong is a very motivating driver for executive decisions.
it shows people are still dumb enough to fall for advertising…
Never underestimate people’s ability to be stupid.