AI coding "assistants"

[s0lar]

Clocking around 20 years of development experience. Here is my read and am I the only one?

My days used to start with requirements gathering i.e. enterprise architecture then I would move on to solution architecture. Diagrams, requirements, dependancies .etc Then onto spooling up the unknown in the stack or SDK or whatever it may be. Taking great pride in my code skills and expansive knowledge across the stack. Delivering a solution with a great sense of pride as it was created top to bottom.

Past couple of month I have been pretty much peer reviewing my prompts to copilot and cursor. In all honesty more often than not the code it spits out is better than mine.

I am not enjoying development anymore. Wheres the challenge? Where is the creativity and creation?

Am I the only one to embrace AI assisted coding but finding the 9-5 now a bit of a bore?

 

[Analog Revenant]

Yeah, same here, been programming since 2000. Now I hardly do any programming. I just manage AI to generate what I need. It's like working with a remote developer communication via a messenger.

Feeling a bit sad, but it has sped up development time for our small company in such a way we can never go back. I used to outsource frontend work (I'm really bad at design and frontend), now AI does all of this for us @ a fraction of the time and cost.

If we dont get the output we think works, we get a UX designer for specific things and just feed the Figmas to AI. Problem solved.

I was an early adopter by using ChatGPT early 2023 already to generate code in Flutter (at the time I had no clue about flutter), but AI has gotten so much better these days!

 

[Analog Revenant]

AI did take something from me though, I used to have the urge to learn, and to learn more languages & frameworks. Now I feel like what is the point?

 

[Kosmik]

Hmmm, I dunno about that. haven't used items like codex etc. for repos but have been developing with AI but the broader the architecture , the more it you need to guide it otherwise it tends to encapsulate a lot vs refactor into other classes.

It eliminates a huge amount of tedious code though but you still need to apply design and principle.

Thats my current experiance. Also, if you are coding "closer to the edge" , as in newer AI frameworks, it lags which makes sense. General code, constructs and older it works well.

Peer coding is a good way to see it I think. Or at least as running a small team.

 

[s0lar]

AI did take something from me though, I used to have the urge to learn, and to learn more languages & frameworks. Now I feel like what is the point?

Exactly, there is no challenge or reason push the boundaries. The increase in delivery velocity is going to be an expectation in the near future. In the beginning using the assistants meant I could spend more time on a design or get to my emails or whatnot and not leave them for tomorrow.

Right now am focusing on my soft-skills and steering my career that way as I dont want to throw the experience accumulated over the years away. Not sure how long that is going to last before its all AI driven. I was in a large semi-technical meeting of 200 odd attendees, it was so obvious one of the presentations was just him reading a fluffy chatgpt created script.

I mean just have the autogenerated teams meeting minutes of a potential product or scope, feed it to GPT as a prompt and have it document, standup up the infrastructure, code and deploy to a review environment.

 

[Analog Revenant]

Hmmm, I dunno about that. haven't used items like codex etc. for repos but have been developing with AI but the broader the architecture , the more it you need to guide it otherwise it tends to encapsulate a lot vs refactor into other classes.

It eliminates a huge amount of tedious code though but you still need to apply design and principle.

Thats my current experiance. Also, if you are coding "closer to the edge" , as in newer AI frameworks, it lags which makes sense. General code, constructs and older it works well.

Peer coding is a good way to see it I think. Or at least as running a small team.

yeah, for sure - good points.

Most of the design principles I add in context files, the more context you give it the better the outcome. But I do admin my current code bases are relatively small - mostly mobile apps with microservices backends etc. It's insane how easy for example Claude Claude keeps track of everything.

I'm on the Anthropic Max plan though - so context window in terms of input tokens and output tokens are way more than on the Pro plans.

 

[SauRoNZA]

I’m on the other side of this again.

I’ve never been a developer/coder but do a lot of scripts and infrastructure as code stuff which usually I wouldn’t even need to write from scratch but instead augment existing library of stuff to fit a new use case.

Now I just throw it at the AI instead and it frees up my time to do other things instead, so I feel less time pressure on myself when I have more meetings than usual and that kind of stuff.

Also helps a hell of a lot when I need to fix something someone else did or something which has become outdated and broken in the process.

 

[Analog Revenant]

Exactly, there is no challenge or reason push the boundaries. The increase in delivery velocity is going to be an expectation in the near future. In the beginning using the assistants meant I could spend more time on a design or get to my emails or whatnot and not leave them for tomorrow.

Right now am focusing on my soft-skills and steering my career that way as I dont want to throw the experience accumulated over the years away. Not sure how long that is going to last before its all AI driven. I was in a large semi-technical meeting of 200 odd attendees, it was so obvious one of the presentations was just him reading a fluffy chatgpt created script.

I mean just have the autogenerated teams meeting minutes of a potential product or scope, feed it to GPT as a prompt and have it document, standup up the infrastructure, code and deploy to a review environment.

Anthropic's leadership has been quite vocal on possible future outcomes in terms of jobs etc. I think the important part will be to keep up with all the models in terms of use and what it can achieve for you.

Anthropic also posted a new ad yesterday, where they do try to spin that you should still focus on solving problems, learning. But with AI:

https://twitter.com/x/status/1968705632095158393

In the end I think it's going to be important to identify when and how to use AI or not for certain things.

So I guess we should be getting in the mindset (I've been trying to get better with this) that we could be tackling any difficult problem now, because we have this massive new technology which is insane if I think how slow our projects used to be pre 2023.

 

[Analog Revenant]

I’m on the other side of this again.

I’ve never been a developer/coder but do a lot of scripts and infrastructure as code stuff which usually I wouldn’t even need to write from scratch but instead augment existing library of stuff to fit a new use case.

Now I just throw it at the AI instead and it frees up my time to do other things instead, so I feel less time pressure on myself when I have more meetings than usual and that kind of stuff.

Also helps a hell of a lot when I need to fix something someone else did or something which has become outdated and broken in the process.

Yes! For me the cool thing has been troubleshooting. I've spent hours and hours before AI on stack overflow and random sites for solving certain issues from compiling problems to firewall configs on my ubuntu servers.

Now I copy and paste errors and problems in Code Claude, most of the time I don't even explain, just paste the error and hit enter.

 

[SauRoNZA]

Anthropic's leadership has been quite vocal on possible future outcomes in terms of jobs etc. I think the important part will be to keep up with all the models in terms of use and what it can achieve for you.

Anthropic also posted a new ad yesterday, where they do try to spin that you should still focus on solving problems, learning. But with AI:

https://twitter.com/x/status/1968705632095158393

In the end I think it's going to be important to identify when and how to use AI or not for certain things.

So I guess we should be getting in the mindset (I've been trying to get better with this) that we could be tackling any difficult problem now, because we have this massive new technology which is insane if I think how slow our projects used to be pre 2023.

It’s not always about difficult problems so much as just removing inefficiencies.

I recently built a simple automation workflow with Gemini in the middle to process invoices and strip out the vendor information, serial numbers, descriptions etc and pump them directly into our asset management platform.

This isn’t difficult work, it’s just brainless work consuming someone’s time with busy work. Now the guys working working with those assets simply chuck them into the workflow and do a quick visual confirmation and save themselves an hour or two a couple of times a month.

The net benefit is having handed it over to them now they’ve learnt a new skill in the process of seeing how this works and their mindset has changed around what they can do for themselves and a week later you find them building something else where they spot a gap for it and all of sudden your level one stock controller guy become a level two automation specialist.

 

[Analog Revenant]

It’s not always about difficult problems so much as just removing inefficiencies.

I recently built a simple automation workflow with Gemini in the middle to process invoices and strip out the vendor information, serial numbers, descriptions etc and pump them directly into our asset management platform.

This isn’t difficult work, it’s just brainless work consuming someone’s time with busy work. Now the guys working working with those assets simply chuck them into the workflow and do a quick visual confirmation and save themselves an hour or two a couple of times a month.

The net benefit is having handed it over to them now they’ve learnt a new skill in the process of seeing how this works and their mindset has changed around what they can do for themselves and a week later you find them building something else where they spot a gap for it and all of sudden your level one stock controller guy become a level two automation specialist.

Good poInt, I hear you, but sometimes there is a possible (probable) bad side to this.

I used to do python automation (Robocorp at the time - now sema4.ai) as a freelancer. We'd pull into a company (mid size) and automate anything we could find in the companies' operations, freeing up resources. Felt great, until I realized later we were in many cases taking away some old aunty's job in the end. That was the end game for a lot of companies - more automation, saving money.

I guess it's been the same since the Industrial Revolution - someone's job always gets taken away at some point in history. I stopped doing that though, I couldn't anymore.

Now AI is accelerating it, as you say.

 

[SauRoNZA]

Good poInt, I hear you, but sometimes there is a possible (probable) bad side to this.

I used to do python automation (Robocorp at the time - now sema4.ai) as a freelancer. We'd pull into a company (mid size) and automate anything we could find in the companies' operations, freeing up resources. Felt great, until I realized later we were in many cases taking away some old aunty's job in the end. That was the end game for a lot of companies - more automation, saving money.

I guess it's been the same since the Industrial Revolution - someone's job always gets taken away at some point in history. I stopped doing that though, I couldn't anymore.

Now AI is accelerating it, as you say.

I mean consulting like that to a company by its very nature is always going to have casualties…but odds are that old auntie was raking in way too much money for doing much too little just because of being there forever and a day and she’s not going to learn new skills and would get there eventually either way.

On the other hand it makes room for youngsters starting at the bottom and gives them a development path and a means to go up in the world.

Not the case in all industries and not a universal truth for sure, but it’s not all bad everywhere.

Maybe somewhat unrelated but I’ve changed my hiring practises the last two years or so and stopped trying to replace seniors from outside, who can’t walk in the door and just start working right away and need six months to become truly useful, in favour of slotting folks in at the bottom instead who are already familiar with the surroundings and upskilling them over time instead.

Means I can get away cheaper for sure, but also means I have an entire promotion path for them over the next few years and they also tend to stick around far longer as there isn’t an automatic ceiling where they have to move to another team or outside to a different role entirely.

And in those cases I think AI stands to benefit them greatly.

 

[DrJohnZoidberg]

Wheres the challenge? Where is the creativity and creation?

I'm just using co-pilot currently but I don't find these tools particularly useful when designing new features as the code base I work on is pretty big and complex. I still have to design everything, I sometimes use Claude to bounce different ideas off but I'm still building the architecture of the overall thing at the end.

I use co-pilot as a glorified auto complete at the end of the day.

I have used Claude Code to build entirely new projects for fast prototyping but the main project I work on is too big for these tools to have enough context on.

 

[Fulcrum29]

I am not a developer at all, but I have done scripting ever since I was a kid. Working with systems like CMS, CRM, some ERPs, Asterisk, etc, you tend to learn a lot as time progresses. Now, I work a lot with microcontrollers, and sometimes I struggle with some more advanced systems in regard to coding. I started to use AI, and well, it works. I will tell you now that each one I have tested isn't 100% reliable. Sometimes it reverses things, so I would rather look to merging. Let's say that had I not checked the other day, an Arduino would have blown out my USB. It would change some lines outright without being instructed. It can also be an issue when using things like stepper motors, relays, etc. Do double-check your code!

For my own things, I use ML Studio + Qwen3-Coder + Void.

OT:

For agentic AI, it is great, but the other day, again, I read a meddled-with report that was tainted by AI. Things aren't being checked by the humans. I also did some summarization the other day, and some outputs are hallucinating, or is a total reinterpration or rephrasing.

Then I also use Agentic AI to do SEO, GEO, etc. All I am going to say is that there are errors that I have to manually resolve.

On the one hand, it saves time, and on the other, it can create problems that need time to solve.

Otherwise, steady as she goes. Still though, sad to see some expecting a big paycheck by thinking that AI can do all the work.

 

[Fulcrum29]

I'm just using co-pilot currently but I don't find these tools particularly useful when designing new features as the code base I work on is pretty big and complex. I still have to design everything, I sometimes use Claude to bounce different ideas off but I'm still building the architecture of the overall thing at the end.

I use co-pilot as a glorified auto complete at the end of the day.

I have used Claude Code to build entirely new projects for fast prototyping but the main project I work on is too big for these tools to have enough context on.

Yeah, that can be an issue. I hope that Ryzen AI processors roll out to consumer motherboards. I want to build a localized AI workstation, and I am limited to 20GB vRAM and more or less 60GB RAM that I can use to partially load, but that is too slow.

Again, this was experimental, but I convinced CoPilot to help me build a CS cheat. I did use a readily available cheat that is on GitHub to extend. For educational purposes, CoPilot did well. However, it lacks on the agentic side. Though I have tested it in VSCode, and with my limited use, anyone can produce positive results, though it would need to be reviewed.

 

[saor]

Past couple of month I have been pretty much peer reviewing my prompts to copilot and cursor. In all honesty more often than not the code it spits out is better than mine.

I despise the prospect of human creativity (d)evolving into the art of getting really really good at prompting AI. I'm lucky in that I use AI tools for certain things, but the overwhelming majority of the work I do is still a human clicking a mouse and work that I enjoy. The day the majority of my day is prompting AI feels like the day I'll start staring out the window more.

AI has jaded me in certain ways. I see digital art nowadays and usually just skim right over it - being unable to tell whether a human spent 50 hrs meticulously crafting this thing or 2 minutes prompting an AI. I do understand how some people don't care and merely enjoy the end-product, but I love the behind the scene details of things almost more than the thing; the time, effort and craft that goes into making something. And when I start seeing the things I do through that same jaded lens...ja, not looking forward to that day arriving.

 

[Fulcrum29]

I despise the prospect of human creativity (d)evolving into the art of getting really really good at prompting AI. I'm lucky in that I use AI tools for certain things, but the overwhelming majority of the work I do is still a human clicking a mouse and work that I enjoy. The day the majority of my day is prompting AI feels like the day I'll start staring out the window more.

AI has jaded me in certain ways. I see digital art nowadays and usually just skim right over it - being unable to tell whether a human spent 50 hrs meticulously crafting this thing or 2 minutes prompting an AI. I do understand how some people don't care and merely enjoy the end-product, but I love the behind the scene details of things almost more than the thing; the time, effort and craft that goes into making something. And when I start seeing the things I do through that same jaded lens...ja, not looking forward to that day arriving.

I appreciate AI as an assistant, and helping me do things that I wouldn't have been able to do without others. Yes, original work does come across, or communicate, as more authentic.

The part I like the most about AI is computer vision, and I must admit I love tinkering with Vision AI. This is an area where AI is more than a mere assistant, but a viable tool. There is still human tuning and assessment involved. It is really cool, but also not something new. This is older than OpenAI's rise. It is super cool programming with computer vision, and it can be applied to anything. This is with what I am playing with when I don't have work coming in

 

[Herr der Verboten]

Clocking around 20 years of development experience. Here is my read and am I the only one?

do you want a cloak and cigar with that? 20 years doesn't always translate into good practices

 

[Herr der Verboten]

Now I just throw it at the AI instead and it frees up my time to do other things instead, so I feel less time pressure on myself when I have more meetings than usual and that kind of stuff.

Really a good tool to do the mindless monkey work. I use chat/grok alot for that, hell they can even do decent markdowns.

Sidebar: I don't use copilot its been really kak so far to be honest; maybe things hav improved, but on inception it was bad... really bad and did not give you the power and speed of 10 developers as most "corps/managers/ceos" climaxed on the golf field in group hugs.

 

[Analog Revenant]

Really a good tool to do the mindless monkey work. I use chat/grok alot for that, hell they can even do decent markdowns.

Sidebar: I don't use copilot its been really kak so far to be honest; maybe things hav improved, but on inception it was bad... really bad and did not give you the power and speed of 10 developers as most "corps/managers/ceos" climaxed on the golf field in group hugs.

agree on your sidebar - copilot sucks