VW to fire top executives at software unit Cariad, source says

Yeah I’ve owned VWs for years and have always been basically happy with the cars’ software. We recently bought an EV, and even though the ID.4 checked pretty much all of our boxes, we skipped it entirely based on the steady drumbeat of software complaints.

I’ve wondered how much of this is a reaction to decaying road infrastructure. In the northeast, I’d be skeptical of sedan durability unless it had 6 inches of clearance. At that point, you might as well make an suv.

My observation here in the UK is there may be a shift back to estates, or that may just be confirmation bias (having done just that).

I wouldn't say tesla's software practices should be emulated. Pushing random updates to prod OTA is terrible practice and look at the number of times cars have rebooted while driving

As someone working in German big companies this is EXACTLY how things are being done, and 10 years later we are still doing things the same and fighting the same old problems (like having 17 different versions of the same software on customers installation).

This reminds me of how practically all BaFin license requirements have this personnel requirement i.e. management must have three years of academic experience and three years practical experience. In other words, the only people who get to start a new payment service provider (PSP) must have already worked in management at an existing PSP, essentially pulling up the ladder.

This, but SAFe

This sounds like a very solid bet to me as well

> IF you are a company with a CRITICAL software deficit, then I hate to tell you this, you probably need to spend money on good developers. Experienced, good developers.

I can assure you that VW has great developers. But developers only develop, there’s a whole layer of people who decide what and how it is developed.

In a lot of "traditional" industries, because you "can't touch" or put it into inventory software is not seen as important "it is just there at the end". Furthermore there is a huge disconnect, because they don't understand software... everything "can later be fixed in software", the procurement department is incentivized to buy the cheapest hardware/chips they can get and you as a software developer are never asked if the requirements can be implemented on the hardware they procure leading to delays because you have to fit all the promised features into a shitty 5cent cheaper hardware and if it fails it's the software departments fault (looks like this is a bit the case in CARIAD when reading some german forums/articles).

The car industry reminds me of the construction industry. It's rife with decades-old institutions that are indifferent to quick changes in an ever software-dominated environment.

The thinking is, "if it ain't broken, then it ain't worth fixing." Unfortunately, recent advances by autonomous vehicle companies and EV makers have shifted the paradigm from a hardware-focus to a software one. I think construction is also headed the same path as stick-built buildings are increasingly on the way out and more modular preassembled manufacturing is coming in.

It's good news for software companies and bad news for traditional construction and traditional auto.

I think an analogy is "enterprise software". You always get the worst software when it's designed and approved by a committee that doesn't get hard user feedback (feedback in the sense that people just won't use it if it's bad). So things are released with shit usability. Add on the cost overruns and scope problems that come from people that don't really know software dictating top down what they think it should do, and you get what happens in most industries.

>Suffice it to say that the gap between programmers and non-programmers in how they think and understand the process of developing software is enormous, and too large to bridge.

Programmers speak programming languages. Common men speak human languages.

Incidentally, this is also why "AI" became such a big hit with the masses in such a short period of time: For the first time ever, common men can communicate with computers using human languages instead of programming languages.

> I believe it's because of the top-down approach to software development. The amount of waterfall in those projects is insane

It's more about consensus driven decision making and waiting to make a decision until all information is present : risk avoidance behaviour. You mix these two and it's a great recipe for disaster.

As shown by Tesla being the boggest car maker by now. Oh, wait...

> Structural software impacts both operation of the vehicle as well as “infotainment”.

From my limited insight, it seems that they are tightly coupling these two together: operational embedded code and the infotainment layer.

So they cannot roll out a UX update without also updating the embedded code that steers the vents.

Not just German ones...

Generally, for a long time they treated components with software the same as any other component: You buy it as a standalone thing from your suppliers, who make it according to your specifications, e.g. lots of the electronics would come from a "tier-1" like Bosch, who many cases farm out making of different parts to their suppliers, who then contract bits from subsubcontractors, ... and in the end everything gets plugged together. The car maker at the top of the pyramid is giving some goals/designs and tests if everything fits together and shouts down the chain if it doesn't. And for physical parts and contained software that does kind of work. It involves a lot of specification writing and -checking, but that's what mech eng orgs are good at. A simple digital instrument cluster isn't that different from an analogue one in that regard, in the end it is "hardware has to have this shape, fulfill this spec, display has to read the following CAN signals to show what's needed". Similarly for a navigation system or software on a engine control unit. Where exactly the lines between what was done in-house and what is contracted fell varied (e.g. some at least nominally controlled the software platform for e.g. the navigation/entertainment unit themselves, others bought that too from the Tier-1 and ordered design customization as needed), but in general the car company is mostly doing "managing" and coordination.

But this has a bunch of downsides:

If you don't have the expertise in-house, you are beholden to what your suppliers can or are willing to do. If they do bad work, you can shout at them (and car manufacturers are quite good at shouting at suppliers), but you can't fix or potentially not even properly evaluate it. Integration across subcontractor trees is difficult to make work well and leads to even more spec-writing and shifting blame if it doesn't go well. Since the tier-1 is probably also not a software company, the software expertise is somewhere deep down the contractor tree and not properly involved with the high-level decisions.

E.g. a common complaint is why the UI in cars is so sluggish: What do you expect if the people deciding about the hardware didn't have trustworthy input from people actually building the software and the software is a hodge-podge from multiple suppliers forced to work against interfaces one supplier made up without concern if its useful for the others... or even pass rudimentary common-sense tests.

It also gets way worse the more integrated things become. E.g. if you have a separate box for the navigation system (in some cars it's literally a box somewhere you can take out and replace if you get a map upgrade, which gets power and a screen connector from the car), there's little to no software integration needed. In a modern fancy-ish car, the expectation is obviously that this is all integrated in one UI, the navigation can appear in the instrument cluster, map updates should probably be online and not require bringing the car to a dealer, you want one central computer instead of multiple small ones, ...

So to stop all this falling apart, someone needs to own it. The Tier-1 could, but it also doesn't have a clue how and the manufacturers obviously don't want to give those more power, so they try to bring it in-house. But if you've never had a internal culture for software to grow this from, it is really hard to do that.

VW specifically said they'd do it by force, with plans to hire thousands of software people over a few years to build a totally new software division that would own the entire stack from scratch. It's going about as well as one would think something like that would go...

Other manufacturers who noticed this earlier and let it grow slowly are faring better. Some had more experimental "labs" divisions which were closer to the software world, whose expertise now can be used. Some have accepted they have to farm some of it out and e.g. adopt Android Automotive, even though the industry was very hesitant about that initially (because it gives Google all that juicy user data and ownership over the connectivity services the industry really would rather control itself).

The article states that the purpose of this new "structural software" was to enable level 4 autonomy in the future.

Cruise is arguably farther ahead, and is owned by GM. This seems to me more evidence that the EU sucks at software

Not really, limited to selected parts of the Autobahn and the car will throw you under the proverbial bus 10 seconds before it would make a fatal mistake. It’s all just PR.

Not that low, Stellantis' margins are 30% for instance.

No. The pay is mostly ok, the project management is shit. Basically in the automotive industry, you get important in your company when what you produce is a big part of the BOM of the car. Software is $0 in the BOM, so you are treated like shit.

Source: I've worked for the automotive industry for ~20 years, on and off.

This is not the case in Europe from what I’ve heard. SW devs with the right domain knowledge in cars get very well compensated.