The Long Goodbye

[HC_79]

Emad referred to heavily optioned vehicles potentially using all the headroom, but I have to believe that those would be unusual cases. It would seem those would be pretty easy to detect running diagnostics in a lab as well as reading logs. In any event, my DE is relatively trouble free, but is fully loaded as well.

Mine is as well so I think that assumption is being completely blown out of proportion.

 

[hmp10]

All that anyone here can really say for sure is that there are quite a few very happy Gravity owners, and quite a few who are extremely frustrated with the bugs, and Lucid appears to think the problems are serious. Establishing the ratio of trouble-free to troublesome vehicles is probably impossible given the data available to us, and the inherent sampling bias of who decides to come to the forum and who does not. Lucid may have better data on that, but they're unlikely to want to share it. I personally don't think Emad is a dishonest person, although obviously he has quite a bias from his senior leadership position at Lucid, and so I generally trust what he said when he talked to @hmp10.

Agreed.

But to those who think I'm exaggerating the problems we're seeing, I spent a while today at Lucid's request sending them what information I could about exactly when and what issues have occurred. So far I have sent them eight time-and-date-stamped photos and videos of different problems we've had just between March 20 and today . . . and just on one of the Gravities. Some of the problems have occurred in both cars, and most have recurred repeatedly, sometimes almost daily. (And quite a few of them have been experienced by other owners who reported them on this forum, sometimes spontaneously and sometimes in response to my posting about an issue.) There are some other issues that I didn't capture with my camera, either because I was driving alone or it was something not easily documented that way.

As I've posted elsewhere, I have also used this forum as a means to log some of these events so that I could recapture the dates, and I haven't yet begun to scour my old posts to send them to Lucid. Maybe I will, and maybe I won't. I'm just tired of all this, and I'm hoping the next two updates will address many of them in a general cleanup of the bugs.

 

[Chesapeake]

OK thanks to all who’ve written, I’ve learned a great deal. Got my amazing Gravity GT six months ago (a 9/2025 build) and it remains trouble-free at 8,100 miles now. What an amazing car - smooth, quiet, FAST, roomy - the Rolls Royce of station wagons. This car will out-anything any of the six Teslas I had before it including all four S3XY models. Sole exception: Lucid’s ADAS is obviously far from Tesla’s FSD. But the latest software u/g has been an improvement. Yes, I’ve had to change batteries in both key fobs to restore their performance. Aside from that, and one solitary squeaky noise that Lucid Service totally nailed in one visit, I really know nothing about all the undifferentiated “software glitches” people have experienced. This is still a newbie and evidently had growing pains for some owners - but if they’re building Gravity GT’s and T’s as solid as mine these days, then it’s really only the price that’s keeping the market thin. Hopefully, Lucid’s excellence in design and (from my experience) their QC as well will continue into their new smaller models, priced for a wider market.

 

[hmp10]

Mine is as well so I think that assumption is being completely blown out of proportion.

That's a shame if it's so, as I was hoping the ~30% more processing headroom Dlala says they're creating would sweep a lot of our issues away. If they don't arise from a common root cause, it may be a longer slog.

 

[Leftlane]

Agreed.

But to those who think I'm exaggerating the problems we're seeing, I spent a while today at Lucid's request sending them what information I could about exactly when and what issues have occurred. So far I have sent them eight time-and-date-stamped photos and videos of different problems we've had just between March 20 and today . . . and just on one of the Gravities. Some of the problems have occurred in both cars, and most have recurred repeatedly, sometimes almost daily. (And quite a few of them have been experienced by other owners who reported them on this forum, sometimes spontaneously and sometimes in response to my posting about an issue.). There are some other issues that I didn't capture with my camera, either because I was driving alone or it was something not easily documented that way.

As I've posted elsewhere, I have also used this forum as a means to log some of these events so that I could recapture the dates, and I haven't yet begun to scour my old posts to send them to Lucid. Maybe I will, and maybe I won't. I'm just tired of all this, and I'm hoping the next two updates will address many of them in a general cleanup of the bugs.

No one wants a full time job capturing data to provide to the manufacturer to fix problems. You are as patient as Job with this stuff.

 

[hmp10]

No one wants a full time job capturing data to provide to the manufacturer to fix problems. You are as patient as Job with this stuff.

I don't know that I'm that patient.

As @Chesapeake just pointed out, outside of software issues these cars are so damned good at everything else that most days it seems worth the effort to do what I can to get them sorted on the software front.

 

[hmp10]

BTW, I would be willing to post the photos I've taken of some of our issues. I would need some help from someone with the videos, as those taken on my iPhone can't download to the forum.

But I'm not going to do this unless there is interest from others.

 

[tmichael]

I'm a Luddite, so help me here. When Dlala said this, I figured it meant that all the systems were continuously running in the background whether they were being used or not.

Is that not the case?

I assume that Lucid is reading the logs when working on problems, as they have always asked me for specific dates and times when issues have occurred. It's why I often try to photograph glitches so that I have a date and time stamp I can give them. However, sometimes they've told me that the logs are not reflecting a behavior that I have documented by photo or video. Such has been the case with the errant LED strip in the dashboard ambient lighting. The logs have shown nothing at the times I have photos of it clearly malfunctioning. Even the Miami Service Center said they detected a miscalibration in the strip software that was causing the problem.

I've also had instances where Lucid is seeing something that I'm not. For example, one time I couldn't get the car in gear no matter what I tried to do. So I called Customer Service, who told me it was because the driver-side rear passenger door was open. No one had been in the back seat that entire day, although I had run some errands. I told her there was no "open door" warning being displayed in the car, but she said she was seeing on at her end. So I got out, anyway, to check the door. It was securely closed with the handle retracted and perfectly aligned with the surrounding panels, but I opened and closed it again. The car finally went into gear.

My example would be not using DDP - the car was designed (hopefully) to be running DDP, while sensing windshield. wiper needs, while closing a window, while ... - well, you get it.

I have also had instances of having an error and CS not seeing it. That's when I figure it is a glitch, where signals might be sending an error code to the dash, but the computer isn't dealing with anything - no actual error code exists. Once the car's state changes - sleep, opening a door, etc. - the warning light goes off, but the computer (and the log) is still oblivious to any error - because one didn't actually exist, but the warning light was activated. Have that with a FDU light right now. No evidence of it from CS logging in.

 

[socktucker]

This is out of my league, and I am admittedly old-school. Seeking input from up-to-date colleagues on the forum.
But the concept of using virtualization in any latency-sensitive environment horrifies me. Please talk me down off the ledge.
Most the above primarily benefit developers but are less effective in deployment in my limited experience (but I do know some about RTOS and responsive firmware).
Is it possible that Lucid is using docker containers in full VMs like Ubuntu and expecting any kind of reasonable response times to round-robin polling? Yikes.

My understanding is they run VMs on OpenBSD

 

[borski]

This is a bad take.

Since the start of the year, every software company and dev shop has been using AI to help with software development. The reason everybody has jumped onboard is that it *has* proven to increase the speed, whether through security/compliance testing, feature development, or automated testing (note that the job posting mentions these use cases). They're not looking to hire non-engineers vibe-coding interfaces.

I mentioned this in another thread a few months back, but when I think 3.4 rolled out, they actually rolled back or reduced some of the fancy design elements in UX 3.0: the drive modes no longer animate, the seats are static images, the side cameras run smoothly w/o some sort of crop or filter through them. I'd bet that the problem wasn't the quality of each element (since each screen, by itself, was pretty well-built), but that there wasn't enough integration testing, when all the software pieces came together and all of them are stacked on top of each other. Hence, the new architecture and more of this type of full, end-to-end integration testing, and I suspect the 30% perf. improvement is work precisely to make the full stack be more efficient.

I agree completely.

 

[borski]

I have no issues with non-generative AI. However, the generative AI has significant issues.

I have personally watched it forced onto developers who then completely outsource their critical thinking. Even for things that are supposed to be "trivial" like building a website, a plethora of errors and inconsistencies crop up. Given their tech stack is C++ mostly, the coding assistants and agentic AI workflows are particularly ill equipped for the nuance and complexity of the language. The cognitive bias effect of AI's expertise in areas you are unfamiliar with is particularly strong and well documented.

Open source has drowned, and amongst the hype are numerous studies and research demonstrating a marked decline in quality:
2020-2024 analysis of code repos: https://www.gitclear.com/ai_assistant_code_quality_2025_research
MIT study on cognitive impacts of AI usage: https://economics.mit.edu/sites/default/files/2026-02/AI, Human Cognition and Knowledge Collapse 02-20-26.pdf
METR's first and second study comparison: https://metr.org/blog/2026-02-24-uplift-update/ (Note the selection bias heavily cited in the second study and code quality differences)

My personal experience has been witnessing the neutering of engineers of nearly all critical thinking skills and the ability to reason about new areas outside their expertise. Engineers previously capable of figuring out errors by themselves handicapped to the point of inepitude and requiring assistance to resolve issues when AI couldn't figure it out for them.

I would argue it is a reasonable take grounded in the realities of the technology's limits and effects on developers both from third party studies, observation of impacts at a Fortune 200, and personal use.

You can absolutely use AI incorrectly. You’ve provided plenty of examples of how to do just that.

 

[64right1]

From the moment I was told about Lucid in 2016 I became intrigued. The Air prototype broke so many automotive molds at once, and the credentials of the engineering team were so impressive, that I began following everything I could find about the company. When reservations opened, I was near the head of the line for a Dream Edition. When the first pre-production Air hit the new Lucid showroom in Miami I was there for opening day, and seeing and sitting in that display car simply blew me away.

As soon as the Lucid stock moved out of the SPAC phase, I invested enthusiastically and continued to do so for some time – not because I didn’t know it was risky but because I wanted to support Lucid’s stated mission and the technology behind it.

In the decade since the first prototype was revealed I have followed the company assiduously and supported it by buying its products . . . a total of five Lucids, three of which are currently sitting in our garage: an Air Dream Edition Performance and two Gravity Dream Editions.

So where is my head today about this company?

Well, I have written off my stock investment. I’m not angry about that, as I’m a big boy and understood the risk.

And I have reveled in the fun and comfort of driving Lucids on those days when they are mostly functioning as intended.

But the Gravities have finally broken me. They have the power, handling, style, room, comfort, and structural solidity that is – or should be – the hallmark of the brand. But they also have the more regrettable hallmark of woefully troublesome software.

The software improvements that eventually came to the Air and brought it up to acceptable standards of everyday reliability have sadly eluded the Gravity, and they don’t appear to be anywhere on the horizon.

We’re going to keep the three Lucids we have for a while yet, mostly because I have put close to $20,000 into each car for radar detection systems and PPF protection. And I will keep posting of our experiences with driving and servicing them as they unfold.

Beyond that, though, the exhaustion of dealing with unrelenting Gravity software glitches – even reaching to opening up the cars and starting a drive – has finally worn me down.

We will not be buying another Lucid product in our household as the Gravities have proved to be too erratic and unreliable for everyday transportation needs, and I'm not one for collecting Garage Queens.

It may soon be a moot point. If Lucid rolls out the Cosmos to the mass market with anything like the software problems of the Gravity, the company won’t be around to sell anything, anyway.

Awesome post; thanks ,, and sorry for your loss(es)

 

[AZLucidGuy]

From the moment I was told about Lucid in 2016 I became intrigued. The Air prototype broke so many automotive molds at once, and the credentials of the engineering team were so impressive, that I began following everything I could find about the company. When reservations opened, I was near the head of the line for a Dream Edition. When the first pre-production Air hit the new Lucid showroom in Miami I was there for opening day, and seeing and sitting in that display car simply blew me away.

As soon as the Lucid stock moved out of the SPAC phase, I invested enthusiastically and continued to do so for some time – not because I didn’t know it was risky but because I wanted to support Lucid’s stated mission and the technology behind it.

In the decade since the first prototype was revealed I have followed the company assiduously and supported it by buying its products . . . a total of five Lucids, three of which are currently sitting in our garage: an Air Dream Edition Performance and two Gravity Dream Editions.

So where is my head today about this company?

Well, I have written off my stock investment. I’m not angry about that, as I’m a big boy and understood the risk.

And I have reveled in the fun and comfort of driving Lucids on those days when they are mostly functioning as intended.

But the Gravities have finally broken me. They have the power, handling, style, room, comfort, and structural solidity that is – or should be – the hallmark of the brand. But they also have the more regrettable hallmark of woefully troublesome software.

The software improvements that eventually came to the Air and brought it up to acceptable standards of everyday reliability have sadly eluded the Gravity, and they don’t appear to be anywhere on the horizon.

We’re going to keep the three Lucids we have for a while yet, mostly because I have put close to $20,000 into each car for radar detection systems and PPF protection. And I will keep posting of our experiences with driving and servicing them as they unfold.

Beyond that, though, the exhaustion of dealing with unrelenting Gravity software glitches – even reaching to opening up the cars and starting a drive – has finally worn me down.

We will not be buying another Lucid product in our household as the Gravities have proved to be too erratic and unreliable for everyday transportation needs, and I'm not one for collecting Garage Queens.

It may soon be a moot point. If Lucid rolls out the Cosmos to the mass market with anything like the software problems of the Gravity, the company won’t be around to sell anything, anyway.

Obviously, as a long time Lucid bull myself, I am saddened to learn that your frustration reached the breaking point. I applaud you for owning three Lucid vehicles. Hopefully, Lucid will EARN your business in the future. While I love my Air GT which has been my daily driver for 4+ years now, I have avoided buying a Gravity because of all of the negative banter on this forum. I am shocked that Lucid dropped the software ball on their second model. I remain optimistic that Lucid will "get it" and enact the needed fixes -- even if too late for you.

I concur with other posts in this thread that Lucid MUST get the v1.0 of Cosmos UI software RIGHT. Day one expectations should be PAAK, reliable key fobs. It has to just "work" every time. Of note, I believe that Cosmos will have Tesla model 3/Y like non-powered door handles. Lucid must get Cosmos/Earth right from day one. It's a must.

 

[borski]

I'm of the opinion that removing the battery on the fob is the only way to do a reset of the fob software

It is.

 

[borski]

I have a fully optioned Gravity (non DE) and do not have anywhere near the number of issues as HMP10. No, my GGT is not problem free but issues are more in the mildly irritating. I would be willing to bet that my experience is closer to the norm for most owners.

I have a DE and am also not seeing the same problems. We haven’t turned anything off except the HUD.

 

[borski]

This is out of my league, and I am admittedly old-school. Seeking input from up-to-date colleagues on the forum.
But the concept of using virtualization in any latency-sensitive environment horrifies me. Please talk me down off the ledge.
Most the above primarily benefit developers but are less effective in deployment in my limited experience (but I do know some about RTOS and responsive firmware).
Is it possible that Lucid is using docker containers in full VMs like Ubuntu and expecting any kind of reasonable response times to round-robin polling? Yikes.

QNX® Hypervisor

The QNX Hypervisor allows you to run multiple OSs on a target system so you can separate critical and non-critical functions, support a wide variety of applications, and reduce hardware costs.

qnx.com

 

[borski]

Agreed.

But to those who think I'm exaggerating the problems we're seeing, I spent a while today at Lucid's request sending them what information I could about exactly when and what issues have occurred. So far I have sent them eight time-and-date-stamped photos and videos of different problems we've had just between March 20 and today . . . and just on one of the Gravities. Some of the problems have occurred in both cars, and most have recurred repeatedly, sometimes almost daily. (And quite a few of them have been experienced by other owners who reported them on this forum, sometimes spontaneously and sometimes in response to my posting about an issue.) There are some other issues that I didn't capture with my camera, either because I was driving alone or it was something not easily documented that way.

As I've posted elsewhere, I have also used this forum as a means to log some of these events so that I could recapture the dates, and I haven't yet begun to scour my old posts to send them to Lucid. Maybe I will, and maybe I won't. I'm just tired of all this, and I'm hoping the next two updates will address many of them in a general cleanup of the bugs.

To be clear: I don’t think anyone is saying you’re exaggerating.

I think some people are saying that others (and maybe you) are *extrapolating* from your evidence to “the fleet,” which may not be correct.

I agree we don’t know the numbers / split.

But I don’t think anyone means you’re exaggerating the issues you have.

 

[borski]

My understanding is they run VMs on OpenBSD

QNX.

 

[allenmhc]

This is out of my league, and I am admittedly old-school. Seeking input from up-to-date colleagues on the forum.
But the concept of using virtualization in any latency-sensitive environment horrifies me. Please talk me down off the ledge.
Most the above primarily benefit developers but are less effective in deployment in my limited experience (but I do know some about RTOS and responsive firmware).
Is it possible that Lucid is using docker containers in full VMs like Ubuntu and expecting any kind of reasonable response times to round-robin polling? Yikes.

So it's not quite right that you have virtualizations for everything in the car. Pretty much all cars nowadays have two separate OSs running:

• A real-time OS like QNX that guarantees execution at a given time interval (i.e., no system interrupts delaying execution) which is used for core vehicle functionality like traction control, brakes, etc.
• A more general-purpose OS like Android Automotive or Linux which runs the infotainment systems, including parts of the Gravity's 2 screens

The two systems are deliberately isolated from each other so that the infotainment system crashing, or lagging, shouldn't impact the car's ability to drive. And in fact, you can test this: if you soft/hard reset the infotainment systems, the car still drives and responds to the steering wheel, accelerator/brakes, lights still function, etc.

Most of what the lagging, glitches, etc. posted here, the rearchitected system, virtualization, etc. all refer to the general OS and not the real-time OS. It's necessary b/c it's not realistic to program the UI/UX or most of the features at a hardware level; the graphics, animations, etc. leverage graphical frameworks, and even if you could program all of it at a low level, you still need to account for differences in hardware manufacturers and vendors, e.g., sourcing ultra-wideband radios for keyfobs and PaaK. Evidently Lucid built their stack at too high of a level and didn't consider the fully-integrated performance of the entire OS stack, and they're trying to rectify that mistake now.

 

[prgarg]

So it's not quite right that you have virtualizations for everything in the car. Pretty much all cars nowadays have two separate OSs running:

• A real-time OS like QNX that guarantees execution at a given time interval (i.e., no system interrupts delaying execution) which is used for core vehicle functionality like traction control, brakes, etc.
• A more general-purpose OS like Android Automotive or Linux which runs the infotainment systems, including parts of the Gravity's 2 screens

The two systems are deliberately isolated from each other so that the infotainment system crashing, or lagging, shouldn't impact the car's ability to drive. And in fact, you can test this: if you soft/hard reset the infotainment systems, the car still drives and responds to the steering wheel, accelerator/brakes, lights still function, etc.

Most of what the lagging, glitches, etc. posted here, the rearchitected system, virtualization, etc. all refer to the general OS and not the real-time OS. It's necessary b/c it's not realistic to program the UI/UX or most of the features at a hardware level; the graphics, animations, etc. leverage graphical frameworks, and even if you could program all of it at a low level, you still need to account for differences in hardware manufacturers and vendors, e.g., sourcing ultra-wideband radios for keyfobs and PaaK. Evidently Lucid built their stack at too high of a level and didn't consider the fully-integrated performance of the entire OS stack, and they're trying to rectify that mistake now.

Very clearly explained, thank you