i have no idea what this means.
1 Like
Simple, when learning Java for example there are many things you should do in best preactice but you will, as a learner, do it different. Also the most, not all AI helper are doing not everything like it should be. So you may consdider that you do stuffs in production which are not a good idea. But you will find out and correct them. In my oppinion it is never a good Idea this “learning on the Job” especially not with programming languages. Cause at least the basics of the language should be there otherwise it becomes a dangerous tour. Prople are doing this and often calling a concultant like my company after while reaching out to points it can’t be done in this wise anymore. Do what you want but be aware of stuffs which can become more complex and more buggy than they should become. Not more not less. When learning from the beginning in production it can’t be a good Idea at all. If you only need to learn a few parts while the fundamental knowledge is there it will be a better start at all.
why do you think projects get refactored? no matter what you do, no matter how experienced you are, and no matter how mature the technologies are, refactoring is a fact of every project, as it moves, as the organization moves, and as the devs move.
even our erp software, which is fifty years old, being maintained by greybeards who refuse to retire, gets wildly reworked. it looks nothing like it did, even a decade ago. first, before they knew any better, back when hierarchal databases were state-of-the-art, they used the search keys as primary keys. then they got normal-form religion, and switched to integers. then they got web religion, and switched to uuid’s. then they decided that tree rebalancing was too intense, so they switched to uuid-7’s.
then they all grew tired, asked themselves why they were making their lives so hard, and every query so complicated, and switched to hierarchial db’s, and search keys as primary keys. they wrote some triggers, and now they play battlefront II, all day.
using an llm to help you get moving doesn’t change the refactoring, even a little. if anything, it reduces the number of bad habits you pick up, at the beginning (“what IS boyce-codd normal form?”), because you are copying the work of others who came before and are way past beginner.
we are definitely seeing awful code from secretaries. which is great. they had an idea, they asked claude for help, and they have a prototype. their boss loves the idea, but none of the models can help them proceed. they reach out. the greybeards are in the middle of hacking the arms off of some poor droid, and force-pushing a star destroyer, so they ask the still-have-hairs to fix the project, so they go back to freebasing spice and mainlining blue milk.
the curve is different. that’s it. no code, low code, and llm’s are all just ways to steepen the torque curve on the front, and flatten it, after that.
Let me say: in medical development and aviation, our development areas it is not possible to act like that. So I guess it depends on the environment.
Says somebody which now got that ai freaks out from time to time. Yey
TBH these tech giants I put no faith in their predictions
When was Tesla going to have full self driving mode ?
Wasnt that supposed to be 8 years ago ?
And on Mars by next year ?
Over my many years I’ve come to learn you need to take what they say and push it out by a decade
Sometimes more
1 Like
I trust Sam Altman even less than I trust Elon Musk, if that’s possible.
(Wait a sec, that’s definitely not possible. I guess I’ll just say nobody should trust either of them, at all.)
6 Likes
Hmm, yes, they are in their own class. But GP has a place directly after. But what ever, programming is a war against the tech giants and there is no difference between all of them. Saying that I realized that I trusted in Xojo for a long time. That was a big mistake. And I thought I will not do that again. Now the MAUI Story. What shall I say? The stupidity behind is that all of us are relying on tools. And people which have the power to change what they want will: change what they want. That’s the point why I love Java (cause it is open source and maintained by many) and Linux. Sadly Linux has the problem that there are many vendors which are not writing Software for it. I guess and hope that this will change one day. Until that day I will need to rely on macOS for development. 
Various Chinese cities have self-driving taxis so it exists even if Musky has failed to deliver.
Somebody asking AI to write Java code could be the trigger event that causes it to wipeout all humanity. Let’s not expose it to fugliness that could make it hate us 
I don’t know if it would not even be better if the mankind would be wiped away from earth …
There are trillions of $ at stake so these clowns are all trying to create an Air of Inevitability around AI (ML really) and I think a lot of it is just to sustain the game of musical chairs around chasing after AGI, building enormous data centers with bazillions of dollars of other people’s money. They know there will be no consequences when the music stops playing, and if they manage to succeed even a little, they will in fact be rewarded.
Some senior software architect published on his blog the other day that he has decided to quit worrying and embrace the coding bots. He paints a picture of himself giving specs to the bot, doing reading or blogging or playing with the dog while it works, then testing, asking for corrections and improvements, go do something else while it works, rinse and repeat. He has come to love this way of working and sees it as being the senior dev to a bright intern and while he doesn’t claim it saves time, it seems to be a lower stress way of working.
But I wonder what it’s costing him over time as increasingly he’ll be out of touch with the frontline nuts and bolts and quirks of what he’s working on.
There was a great photo of some utility pole worker working on a tangled nest of wires, probably somewhere in Asia I suppose … the title was “when vibe coding becomes vibe debugging”. I think that will happen at some point too, even to an experienced dev.
I’ve generally worked solo in my career, and have never embraced TDD. I have some unit test coverage, some integration tests, but I’m definitely not a “test first” kinda guy, I’m more old school, cut my teeth on Steve McConnel’s Code Complete and believe in “step through all new code in the debugger”. But if I were going to try the CodeBot thing I think TDD would be a must. Write the tests first, give the bot the requirements, require it to pass tests; when it doesn’t, have the bot DX / fix each bug.
Maybe that would work. Would it save time? Questionable. Would the quality be as good? Questionable. Would I be learning / growing as much, refining my intuitive sense of the problem space? Almost certainly not.
What this would require to put me over the top would be the ability to truly jump to a higher level of abstraction. That is what put me over the top when SQL came along (vs a proprietary DML like xBase languages such FoxPro provided). VB and its spiritual successors is what put almost everyone over the top to create apps with event-driven GUIs.
In fact VB and WinForms and the like are a almost a canonical example. Sure you tend to lose touch with the underlying rendering engine, but you usually gain more than you lose and the tool vendor has implicitly committed to fix issues you run into at that it can address at that low level – to refine the abstraction to accurately work with the underlying APIs. You can then write complex and user-friendly UIs without an intimate knowledge of how the rendering works; in fact, in the cross-platform world, it becomes a virtue to consider that irrelevant and rely on the tool to adapt to each OS you deploy to. Line-of-business devs can focus on business requirements, which is plenty to have to deal with.
I haven’t yet seen AI provide this sort of watershed transformative value, and I’m dubious that it ever will, across the board. Right now AI is great at pure pattern discovery (reading X rays and MRIs, associating subtle things on a chest X-ray with potential diabetes, stuff like that). In the world of software development, it’s not bad at generating gruntwork / biolerplate code, but it falters even there. When it comes to doing more complex things it can kina-sorta follow a guiding hand (you) but it seems so needy and lacking in true understanding that it’s like an intern that never really learns to reason for themselves – at some point it’s easier to Just Do It Yourself.
2 Likes
It’s the Google “I’m feeling lucky” button on steroids.
1 Like
that may then just make stuff up and fee you that with its air of certainty
1 Like
Ya’ll are missing the point.
AI isn’t just a toy—it’s a way to extend the skills you don’t have. I’m lousy at graphic design, but I built https://www.perilpulse.com in about seven hours. What people now call “vibe coding.” AI handled everything: layout, color palette, structure—hell, even the 404 page.
Look at the source. It’s clean. The API page? Would’ve taken forever by hand or required pulling in extra libraries. I didn’t write a single line myself. I just told Claude what to build. It even generated the robots.txt, sitemap.xml, and organized the file tree like a human dev would.
Point is: stop wasting time banging your head against stuff like regex, Excel formulas, or tedious front-end tasks. Use the tool. Stitch together enough of these “small” tasks, and you can ship real, working software fast.
4 Likes
personally I’d have to do a LOT more testing to make sure the code it generated was something I could understand & maintain to have the kind of trust in it you do
Have you had to use AI to do any maintenance to the code its spit out ?
Could you ?
FWIW I’m curious what the prompts for peril pulse even looked like
The prompts were long conversations dictated using SuperWhisper. Paragraphs long, describing exactly what I wanted I simply allowed it to write the files to disk.
We went through 43 iterations to get there.
But the amazing part is if I made a change to something, it was smart enough to go back and update the sitemap and update any links that it was affecting across the entire folder.
Here are some of the 43 conversations:
https://claude.ai/share/d0a7b289-f092-4d74-a104-f438ef59c911
https://claude.ai/share/4cb9d4d9-feb9-4d98-9844-9c927ee05b5c
https://claude.ai/share/89eb6d61-de98-49dc-8bb1-e22ae55ad282
https://claude.ai/share/fa7f88c4-81d4-497e-9a2c-14ce3b2b1e1c
1 Like
probably need to add to the scripts to make sure it doesnt craft one that has certain well known security vulnerabilities as well
And then how do YOU & your team maintain this ?
Since I cant see the actual code it generated I have no IDE if the code is good, bad, indifferent nor whether its “maintainable” by a person (or AI or what)
Being able to spit something out fast is only one aspect of software
The vast majority of a piece of softwares lifetime is “maintenance and enhancement” whether its a desktop app, mobile app, or web app
And dont get me wrong I’m not dissing I’m just curious about the life time of this sort of software
1 Like
norm, it’s no different than having to maintain the code that you wrote for me. The difference is that the documentation it writes is much more complete.
If the application is small or the LLM’s aspect window is large enough, you can just dump the entire codebase in there and start asking questions. Gemini is up to a million tokens, which is a pretty large project.
Don’t get me wrong, there’s still a huge need for a software architect, but not a code monkey. You still have to come up with the logic and tell it what you want done. But it just does it a whole lot faster than you could do it yourself or a team of people. Now if I can get just consistent outputs in Xojo, I’d be a happy man.
