What's new
By:

Over reliance on AI?

merveille07

merveille07

Active Member

Joined
Jul 1, 2025
Messages
255
Reaction score
382
Disclaimer: This is all speculation.
I was a DevOps professional until the last two years, during which I started shifting closer to the development line. I specialize in Unix OS, and while I can work with Windows, I am not as proficient as I am with Linux.
Lately, I've been writing a significant amount of code for both Windows and Linux to automate operational tasks within my organization. As you might expect, I rely on AI tools heavily. Here are my key takeaways:
AI requires complete guidance: The output is only as good as the specificity of the prompt.
AI isn't good at original problem-solving: While some may disagree, I find you need to thoroughly conceptualize the problem and your desired solutions yourself before AI can provide truly meaningful, helpful assistance.
Coding with AI without expertise is a serious security risk: Using AI to generate code in a language or context you don't fully understand is hazardous on many levels.
Why do I bring this up?
I have been active on this forum for over six months, observing multiple software releases. I suspect the problems we are consistently seeing could stem from a combination of the following issues:
Over-reliance on AI: Teams might be using AI to generate complex features without sufficient peer review or internal validation.
Possible hiring mismatches: I frequently observe professionals who over-represent their skills during interviews but fail to deliver once hired. As a newer company, Lucid may have fallen victim to this.
Poor QA practices: Many users have discussed the QA shortcomings, but few have explicitly connected this to AI's role in the development pipeline.
AI is a powerful tool, but you must have experts in the field vetting anything it helps write or suggests. AI can accelerate development considerably, but it must be used carefully and with a critical human check.
Whatever the root cause, I believe Lucid needs to undertake a serious, structured review of its software development process. The more cars they get on the road, the higher the risk posed by these unresolved software bugs if they are not urgently addressed.
This is only speculation.
 
I wouldn't leap to the conclusion that use of AI has been a problem for Lucid's software development. To me, it's much more likely that the problem is the difficulty of hiring engineers who can build embedded systems that have the functionality richness we more traditionally expect in non-embedded environments. Those are two very different kinds of expertise, and many companies have struggled to combine them. Not to oversimplify, but it's part why VW is paying Rivian $5B.
 
It does make sense what you say tho. Maybe the real answer to why simple features aren’t added (geo fencing auto lock, better audio controls, a phone key that makes sense, etc) is because on some level they don’t know how the code works, and their AI tools can’t spit out what they need it to.
 
socktucker said:
It does make sense what you say tho. Maybe the real answer to why simple features aren’t added (geo fencing auto lock, better audio controls, a phone key that makes sense, etc) is because on some level they don’t know how the code works, and their AI tools can’t spit out what they need it to.
Click to expand...
Even before this every-tech-bro-and-CEO-pushing-AI dystopia we live in now, they still didn't have those features implemented. I'd say it's probably management making different priority choices and lack of development resources... That's usually the limiting factor at my job.

PS if a hiring manager at Lucid is looking for a remote embedded SW architect with c++, Java , and Android experience...
 
merveille07 said:
Disclaimer: This is all speculation.
I was a DevOps professional until the last two years, during which I started shifting closer to the development line. I specialize in Unix OS, and while I can work with Windows, I am not as proficient as I am with Linux.
Lately, I've been writing a significant amount of code for both Windows and Linux to automate operational tasks within my organization. As you might expect, I rely on AI tools heavily. Here are my key takeaways:
AI requires complete guidance: The output is only as good as the specificity of the prompt.
AI isn't good at original problem-solving: While some may disagree, I find you need to thoroughly conceptualize the problem and your desired solutions yourself before AI can provide truly meaningful, helpful assistance.
Coding with AI without expertise is a serious security risk: Using AI to generate code in a language or context you don't fully understand is hazardous on many levels.
Why do I bring this up?
I have been active on this forum for over six months, observing multiple software releases. I suspect the problems we are consistently seeing could stem from a combination of the following issues:
Over-reliance on AI: Teams might be using AI to generate complex features without sufficient peer review or internal validation.
Possible hiring mismatches: I frequently observe professionals who over-represent their skills during interviews but fail to deliver once hired. As a newer company, Lucid may have fallen victim to this.
Poor QA practices: Many users have discussed the QA shortcomings, but few have explicitly connected this to AI's role in the development pipeline.
AI is a powerful tool, but you must have experts in the field vetting anything it helps write or suggests. AI can accelerate development considerably, but it must be used carefully and with a critical human check.
Whatever the root cause, I believe Lucid needs to undertake a serious, structured review of its software development process. The more cars they get on the road, the higher the risk posed by these unresolved software bugs if they are not urgently addressed.
This is only speculation.
Click to expand...
I hear what you're saying, and it's not impossible, but it's unlikely. I strongly suspect the issue is in software management and/or team size, not use of AI; for one thing, the rate of new features and bug fixes has not skyrocketed (and has even slowed at times), which is a good indicator that they haven't suddenly just started pushing out Claude-coded crap. :)
 
borski said:
I hear what you're saying, and it's not impossible, but it's unlikely. I strongly suspect the issue is in software management and/or team size, not use of AI; for one thing, the rate of new features and bug fixes has not skyrocketed (and has even slowed at times), which is a good indicator that they haven't suddenly just started pushing out Claude-coded crap. :)
Click to expand...
I hope so. I am just intrigued by the fact that every single release except the HDFA has brought more headaches than actually solving the issues. I mean you are suppose to test your codes piece by piece, then together and then go through extensive QA aka real world testing in this case; so it is hard to imaging flopping on the same issue for this long. Anyway I just Lucid get this fixed sooner than later.
 
I gave my own analysis here that you may be interested in.

Post in thread 'Motor trend Lucid Gravity review with Kyle Conner' https://lucidowners.com/threads/motor-trend-lucid-gravity-review-with-kyle-conner.13669/post-300751

Those issues seem far more likely to be the root cause than AI slop. I suspect there are at least some engineers relying on "code assistants" as a cop out of engineering laziness, but this should be minimal given the culture of Lucid.

One thing I partially noted in that but is rather apparent as I remember the tech deep dives is the mountain of in-house simulation platforms. Those are very resource intensive with constant requests for new and/or improved functionality. The bulk of my career has been in that space of physical simulations and supporting analytical platform development, and the demands at Lucid are likely higher given the advancements in technology relying on those platforms.
 
Rogue said:
I gave my own analysis here that you may be interested in.

Post in thread 'Motor trend Lucid Gravity review with Kyle Conner' https://lucidowners.com/threads/motor-trend-lucid-gravity-review-with-kyle-conner.13669/post-300751

Those issues seem far more likely to be the root cause than AI slop. I suspect there are at least some engineers relying on "code assistants" as a cop out of engineering laziness, but this should be minimal given the culture of Lucid.

One thing I partially noted in that but is rather apparent as I remember the tech deep dives is the mountain of in-house simulation platforms. Those are very resource intensive with constant requests for new and/or improved functionality. The bulk of my career has been in that space of physical simulations and supporting analytical platform development, and the demands at Lucid are likely higher given the advancements in technology relying on those platforms.
Click to expand...
That was a great analysis. I think you might be onto something.
 
You're spot on about AI requiring complete guidance. It is a great accelerator for boilerplate code and quick automation scripts, but it completely falls apart when it comes to actual architecture and original problem-solving.

With an automotive company like Lucid, this gets dangerous fast. In standard software development, a bug means an app crashes or a UI element breaks. In an EV, bad code can affect actual vehicle operations, thermal management, or brick the infotainment system entirely.

The point about hiring mismatches is a huge issue across the tech industry right now. It is incredibly easy for a candidate to look competent in an interview or complete a take-home assignment by leveraging AI. But once they are hired and have to debug complex systems or understand how code interacts with vehicle hardware, the lack of core expertise shows up immediately.
 
I wouldn’t jump to the conclusion that AI is the problem. Emad said in a recent interview that the software team was siloed and if a business has too many silos, especially within one team it’s a recipe for disaster. From the outside looking in, it seems it’s always been a management problem. They’re on their 3rd (maybe 4th) team overhaul now. Thats not an AI problem.
 
HC_79 said:
I wouldn’t jump to the conclusion that AI is the problem. Emad said in a recent interview that the software team was siloed and if a business has too many silos, especially within one team it’s a recipe for disaster. From the outside looking in, it seems it’s always been a management problem. They’re on their 3rd (maybe 4th) team overhaul now. Thats not an AI problem.
Click to expand...
To be clear; I didn't say AI is the problem. I made hypothetical statement. 😂😎
 
merveille07 said:
To be clear; I didn't say AI is the problem. I made hypothetical statement. 😂😎
Click to expand...
These days I am using AI nearly daily and agree wholeheartedly with your analysis.

For those in executive roles, one issue is the profound certainty with which AI presents a solution, which is unfounded. But the extreme confidence tends to suck in the unknowledgeable. Some of the stuff it has presented to me is beyond terrible.
 
AI didn't invent bad code, but it was sure trained on a lot of it.
 
Hammerthyme said:
AI didn't invent bad code, but it was sure trained on a lot of it.
Click to expand...
This a really great point which I try to hammer home as often as it comes up.

Top code from the brightest is not on the internet, but a lot of students seeking homework help is.
-------------------------------------------------
Returning to car topics, when Tesla announced they were going from a rules-based model for FSD to an observationally trained one, I cringed.

I really don't want my vehicle driving as badly as a lot of the drivers out there.
 
AI is great for some things, terrible for others. The trick in knowing where it excels and where it falls short. It sounds like code is an area where it may fall short much of the time.
 
drb3333 said:
Returning to car topics, when Tesla announced they were going from a rules-based model for FSD to an observationally trained one, I cringed.

I really don't want my vehicle driving as badly as a lot of the drivers out there.
Click to expand...
Somewhat off-topic perhaps, but I laughed out loud when I read an article right after Tesla released their Cybercab in Austin. The article said the cabs were slamming on the brakes whenever near a police car. Immediately I knew what was up and cackled "The ****ing AI learned to panic brake by watching videos of stupid drivers!!"

I don't know whether to call that over-reliance on AI, or over-reliance on real people and their bad behavior. The blind leading the blind...
 
You must log in or register to reply here.
Back
Top