Beyond the Cloud: The Local-First Software Revolution • Brooklyn Zelenka & Julian Wood

Posted by goto-con@reddit | programming | View on Reddit | 11 comments

[-]

binarypie@reddit

I swear tech is one giant timeline wheel... different age same stuff.
It's cool to see this come back into the spotlight the early to mid 2000s will live again... or maybe no because most people don't care about connectivity and the complexity to develop local first apps even with an abstraction is still higher than saying "drive to starbucks" in most business use cases.

[-]

Full-Spectral@reddit

After a decade or two of paradigm X, you have a whole generation of developers who have mostly only ever experienced paradigm X. It has lots of issues and complexities that drive them crazy, so obviously something else must be the answer. And of course talking heads want to be edgy and whatnot and tend to enhance the disillusionment (and they may be correct of course.)

So suddenly everyone moves to paradigm Y and the process starts over again. Initially the issues are different so it must be better, but a decade or two later a generation has grown up that only has used Y and is fed up with its issues and complexities.

It's just like politics flailing from left to right in a continuous cycle. The actual answer, that there is no fixed answer, never wins, so we just keep running around in circles.

Obviously some forward progress gets made, I don't want to be overly cynical.

[-]

MilkFew2273@reddit

The actual answer is there is no one size fits all, because the problem domains are different. We are still looking at software and IT as a quasi-academic problem, trying to make money from it. In reality, software should have been another tool ( or the main tool ) for other disciplines to enhance their workflows. A businessman that can write his own financial analysis stuff. Perhaps that's the real selling point for AI, that you can finally have the computer "do the thing" without that pesky layer. We are still trying to figure out a COBOL for every little domain.. And we still want essentially someone else to understand all the nuance. We want to keep decision making and offload the actual implementation to the craftsman. But the devil is in the details so for any sufficiently complex domain, this will remain an art , even though we tend to just buy a printed poster instead of commission a painting..

[-]

Full-Spectral@reddit

To spin it around, why do software developers go to a business person to get help with their finances if they are complicated? Because it's complicated and it takes a lot of time to understand the issues and being wrong could have consequences. You could use an LLM to fill out your complicated tax returns, but would you actually do that? The IRS isn't going to cut you any slack because you used an AI and it was wrong.

The problem is that LLMs don't actually know anything. They can only regurgitate stuff they've consumed and they have no idea if it's right or not. They are not experts in anything. So, for anything that matters you will have to know enough to know if it's right or wrong, in which case you have to have sufficient knowledge yourself. And then we are right back where we started above.

That's why business people don't write their own code generally, and why LLMs aren't going to change that in any major way.

[-]

MilkFew2273@reddit

Or you can vibe code it until it mostly works and when it doesn't you find a bug. I'm not saying LLM are going to fix that magically I'm saying that in an ideal world the subject matter expert should write the programs. If it's easier for an accountant to write python or a pythonista to learn accounting, you tell me.

[-]