Artificial Intelligence thread

[BlackWindMnt]

As you said, this problem has to do with the stakeholders, nothing to do with AI's ability.


No need to fix code. By then, it would just be cheaper and easier to ask AI to generate new code every time.


These two issues are easier to solve than you might think. I saw a demo that does exactly these — project management. This is not the ChatGPT that you would use via a browser, but a dedicated software interfaces with ChatGPT via API. It also uses up to a million tokens in an hour. The problem lies in translating those requirements into codes. I would actually go as far as saying that project management would get automated first before coding would, because code must follow a very strict formatting standard spanning across multiple files.

You want stability in your code base, you can't generate piles of code every time a bug comes in or a new features is being requested.
What if the newly generated code has new bugs or has business killing performance issues. How will you fix that, just reprompt and pray to rngjesus that everything will be hallucinated well this time. How will users of the end product react when the AI generates completely random user interface after each prompt and deployment. There's a reason why most (semi)professional software teams write almost equal amount of unit, integration, e2e, ui tests code compared to feature code just to make sure the new code doesn't break existing code.

 

[Engineer]

You want stability in your code base, you can't generate piles of code every time a bug comes in or a new features is being requested.
What if the newly generated code has new bugs or has business killing performance issues. How will you fix that, just reprompt and pray to rngjesus that everything will be hallucinated well this time. How will users of the end product react when the AI generates completely random user interface after each prompt and deployment. There's a reason why most (semi)professional software teams write almost equal amount of unit, integration, e2e, ui tests code compared to feature code just to make sure the new code doesn't break existing code.

We already generate piles of code for every bug fix or a new feature. That process is called compiling. Almost no one edit the resulting binaries by hand. Modern day source code will eventually go the way of Assembly, optimized for automation and performance rather than readability, because no one is going to read it.

Also, any AI code generation is going to do things incrementally. So assuming you are not changing requirements, asking AI to update the code should only result in a small part of the code base being regenerated. That is also how compiling works today.

The demo that I saw also outputs code for unit tests, integration tests, deployment, and documentation. Knowing what to generate is the easy part. The difficult part lies with making code consistent across the entire code base.

 

[BlackWindMnt]

We already generate piles of code for every bug fix or a new feature. That process is called compiling. Almost no one edit the resulting binaries by hand. Modern day source code will eventually go the way of Assembly, optimized for automation and performance rather than readability, because no one is going to read it.

Also, any AI code generation is going to do things incrementally. So assuming you are not changing requirements, asking AI to update the code should only result in a small part of the code base being regenerated. That is also how compiling works today.

The demo that I saw also outputs code for unit tests, integration tests, deployment, and documentation. Knowing what to generate is the easy part. The difficult part lies with making code consistent across the entire code base.

I have seen enough AI demo where i don't fear AI taking over software development anytime soon at least not in the next decade. Maybe in 2040s it might happen.

Time will tell what will happen, but if I had to bet I would bet i will still be coding in the 2030s.

 

[4Runner]

I have seen enough AI demo where i don't fear AI taking over software development anytime soon at least not in the next decade. Maybe in 2040s it might happen.

Time will tell what will happen, but if I had to bet I would bet i will still be coding in the 2030s.

AI will incrementally be taking place lots of software jobs in coming years and decades. As of now, general usage is for aiding software engineers in basic or repetitive software snippets in simple functional contexts. I am finding it very helpful and useful in that I rarely go back to my old files nowadays instead I just use a combination of Copilot, AmazonQ, Grok and ChatGPT-4o. I think, at least for the next 5-10 years, it will be helpful to experienced software engineers in deep-knowledge domains but may reduce needs for new grads in general IT.

 

[9dashline]

---

AI will incrementally be taking place lots of software jobs in coming years and decades. As of now, general usage is for aiding software engineers in basic or repetitive software snippets in simple functional contexts. I am finding it very helpful and useful in that I rarely go back to my old files nowadays instead I just use a combination of Copilot, AmazonQ, Grok and ChatGPT-4o. I think, at least for the next 5-10 years, it will be helpful to experienced software engineers in deep-knowledge domains but may reduce needs for new grads in general IT.

Half of all software dev jobs will be gone within 2 years, AI doesnt have to replace 100%, but if it lets one guy do the work of 3, the other 2 are getting laid off, they aint all 3 gonna get a 3x pay raise

Please, Log in or Register to view URLs content!

 

[BlackWindMnt]

AI will incrementally be taking place lots of software jobs in coming years and decades. As of now, general usage is for aiding software engineers in basic or repetitive software snippets in simple functional contexts. I am finding it very helpful and useful in that I rarely go back to my old files nowadays instead I just use a combination of Copilot, AmazonQ, Grok and ChatGPT-4o. I think, at least for the next 5-10 years, it will be helpful to experienced software engineers in deep-knowledge domains but may reduce needs for new grads in general IT.

I could definitely see tooling help developers, I myself would love a more accurate documentation tool that returns links in documentation and code snippets. That can run locally and in the terminal returns an result within 2 or 3 seconds as some sort of replacement for Google and github issues searching...

 

[4Runner]

I could definitely see tooling help developers, I myself would love a more accurate documentation tool that returns links in documentation and code snippets. That can run locally and in the terminal returns an result within 2 or 3 seconds as some sort of replacement for Google and github issues searching...

I agree. An AI documentation tool for source code (files, folders, modules, repos, etc.) would be very valuable to software engineers as well as business units. There is a running joke in my business: everyone wants others to document their software/solution; yet they themselves are slackers in documenting their own software/solution. So RTFSC becomes a standard practice

 

[Wrought]

There is a running joke in my business: everyone wants others to document their software/solution; yet they themselves are slackers in documenting their own software/solution. So RTFSC becomes a standard practice

That's literally the whole industry bro.

 

[BlackWindMnt]

That's literally the whole industry bro.

The whole movement to scrum has made writing documentation a curse don't need to write docs because the code is self documenting...

People wonder after 4~6 years when the original developers, product manager, analysts moved on or put on new projects why no one really knows why business rule 103 was needed or what it actually calculates etc etc. Now that i almost a decade worth of experience and have worked on project longer then most people do on their CS degree, have been put on decades old code bases i started appreciating projects that still have their design docs.

 

[Engineer]

I have seen enough AI demo where i don't fear AI taking over software development anytime soon at least not in the next decade. Maybe in 2040s it might happen.

Time will tell what will happen, but if I had to bet I would bet i will still be coding in the 2030s.

Your timeline only makes sense if progress were linear. The issue is that progress is geometrical. Ten years ago we were still struggling with image classification. Today, image classification is just an introductory to deep neural networks. I will give it at most five years before AI becomes capable of software engineering. 2030 is going to be the watershed moment when big-techs massively layoff their developers. Sure, we would still be programming in the 2030's, but the massive redundancies means we would be doing it at minimal wage instead of six figure salary.

The whole movement to scrum has made writing documentation a curse don't need to write docs because the code is self documenting...

People wonder after 4~6 years when the original developers, product manager, analysts moved on or put on new projects why no one really knows why business rule 103 was needed or what it actually calculates etc etc. Now that i almost a decade worth of experience and have worked on project longer then most people do on their CS degree, have been put on decades old code bases i started appreciating projects that still have their design docs.

Documentation doesn't generate revenues so no priority is given to it by management. Developers are overloaded with impossible deadlines that even writing good code is becoming a problem, forget about sparing time for documentation. Remember when disciplines with memory usage used to be a thing? Peppermint farm remembers. In the West, we have a pandemic of a bunch of know-nothing with arts and business degrees being in the driver seat making policies, then expect those in science and engineering to create miracles. It is a cultural problem rather than a scrum problem.