Get ready for this to become a common theme. Boardrooms are still engaged in the fever-dream promise that AI will solve all their problems, particularly those involving pesky humans. The simple lesson of "AI is another tool" will be a hard-learned one. Some industries, such as software, will take more time to mop themselves into a corner before they discover that velocity should never be a first-class concern. Speed should only come as a side-effect of quality.
You seem like a person who works at a place that doesn't have an AI mandate. That sounds nice. I miss when we had nice things in the world like that. I will never take that for granted again.
AI mandate is one of the best things that's happened to me. It's the easiest metric to game in the world.
At one point my boss asked why my AI usage was lower than other team members. I instantly knew what to do. Every session is now run at ultracode effort. My automated PR review bot averages like $80 in usage per PR review.
It is extremely easy to burn tokens if that is required.
Explore this codebase.
Team x wants y feature, research and generate a full plan.
What does feature x in codebase y actually mean?
Analyze code coverage in x.
Map out code flow and find concurrency bugs in y
and on and on...
Oh and my favorite: Use 5 independent subagents to review code change and summarize the findings, and for any finding determine if they are real concerns
The other day claude spun up 100 agents and took an hour to type 30k token document to tell me something was impossible to do. I googled it, found a pr on the 3rd link that showed it was possible. "You're absolutely right!!"
There is value in doing all that too, though. Admittedly with strong diminishing returns, but it's there.
Eg by doing that I was able to develop non-essential features which increased our quality of life for devs last month without going through our PO who'd need to price it - because that does let's you create changes in an incredibly hands off manner with miniscule amount of time investment if you already know what you want to achieve, and how the end result should be...
Admittedly, that's a pretty narrow usecase which is rarely the case- but if it is...
And the more uselessly amusing thing is that the manager who requests higher tokens usage probably also doesn't care whether it's producing slop or not. Metric goes up; managers happy until CFO is reported income hasn't gone up as quickly as costs, and that makes the CEO optimistically concerned. Never expect underlying thought from a messenger.
It's interesting that LLM barely had any vetting period or experimentation phase. Suddenly everyone was supposed to test it in production, it seems.
It's even worse/better. It's corporate financial malpractice. At some point they will wake up after the AI psychosis dies down. That might take 1-2 more years. After that most companies will realize that AI is a tool, as OP said, and adjust budgets accordingly.
Importantly, "adjusting budgets" here is for most companies, you know the ones you have to fight to even get an IDE license, a euphemism for zeroing the budget.
It's also the easiest way to determine if your management has AI psychosis or not, and make corresponding decisions about whether to stay with the company.
I'd unironically like my workplace to cover AI spend for me.
There's so, so much mechanically simple but time consuming refactoring that should be done but nobody ever does that because there's never enough free time. Or even various utility scripts and at least finding out of date docs (or writing very basic ones where none exist, though it'd be hard to get them not to feel like slop writing). Or figuring out what additional custom linter rules would be useful, how to improve the CI pipelines and so on.
If I had the Anthropic Max 20x subscription, I could make a large part of the technical backlog disappear (relatively safely).
He's just making a general "efficient markets" argument. He's arguing that whatever happens in a couple of years will be the right thing, no matter what is happening now.
That is essentially not an argument in any direction.
It wasn't meant to be a literal statement, more just a reflection that the situation is so bleak that I cannot imagine a better future; anybody expressing even a little bit of it seems to me like a somebody who has not been crushed into compliance through force.
Quoting the host of the recurring Quiz Broadcast sketch from That Mitchell and Webb Look: "Books mention 'hope'. What was 'hope'?"
As we have seem with offshoring, any company whose main business isn't producing software, isn't coming back in-house, even if the quality for engineering team themselves sucks.
No doubt, but the issue I think they keep running into is they don't understand how useful those "human tools" are, so they keep trying to replace the functions humans provide with AI, without realizing all the other functions that the humans also provided.
Marx had a way to think about that. He would distinguish between labour as in generalized socially necessafy labour, and specific skilled labour.
Value is measure in generalized labour, since that the universal measure of human effort. The genealized amount of time a human being must spend to produce something from its parts. Generalized labour is also what's bought from labourers. You don't pay them to do something specific, you pay them to labour in general.
This contrasts against specific labour, which is whats actually required in the moment. Generalized labour power must be the right kind of specific labour to actually produce anything of value.
The AI leaders have been told that AI is labour. To the extent that it currently is, which I believe is only the case because the market hasn't adjusted, it's not the right specific labour to male anything valuable.
Nah, that’s the future executives problem, the current executive gets to brag about how their AI integrations cut costs while maintaining an acceptable yet enshittified quality
Ford has hired 350 engineers over the last 3 years which happened alongside short comings in using AI inspection tooling.
This has nothing to do with LLMs and instead is almost certainly about their MAIVIS and AiTriz pilots, which use old school CNNs on custom IBM hardware to do visual inspections.
Dirt bag media will do anything for your clicks and leave you more uninformed at the other end.
OP to me sounds more authentic and seems to have inside information.
After a quick search I found a publication actually mentioning about these tools:
Ford previously told Business Insider that it had developed two bespoke AI-enhanced scanning tools that helped validate that cars were properly assembled before rolling off the lot. The tools, called AiTriz and MAIVs, both debuted in 2024.https://autos.yahoo.com/policy-and-environment/articles/ford...
And after doing cursory research on these tools, it is clear they are rudimentary (as compared to SOTA LLMs), they were essentially smartphone mounted on stands and doing visual checks using the camera - so OP could be very right.
Nothing in the article contradicts their (IMHO accurate) claim. Three years ago boardrooms were not drinking the LLM Kool-aid yet, while ML-powered QC has been around for years. Remember Silicon Valley's hot dog vs not hot dog? That's pretty much all you need, only the hot dog is a car part.
This doesn't seem like it backfired. Firing these people and rehiring a fraction of them catapulted Ford to the top. In fact, these roles were apparently there for over a decade before modern AI even came to exist and Ford was never top. This actually presents a formula for improved reliability - fire almost everyone, then hire back the cadre with value. A very DOGE-esque approach and I'm surprised it worked.
Back in the nineties Ford ran a lot of ads about how quality was job one. But in the last twenty years their quality declined by a large amount at the same time other brands were getting better. I say that as a lifelong fan of Ford, quality was why I left the brand two years ago.
And yet all the time you spend performing those recalls should be annoying. Maybe you don't plan to eventually sell your car on the second hand market but if you do, a car without all the required recalls could have a lower value than one with all the recalls applied.
eh, every 6 months to a year I bring the car in to the dealer to handle the stack of pending recalls, during which I get a rental, courtesy of Ford. It's not much of a deal for me.
Few of the issues I've experienced with the car were clearly tied to quality issues: 1) Battery died a few times, but maybe that was user error 2) squirrels/rats nibbled the engine cable harness, a not-uncommon occurrence in our area. Only 3) auto-unlock on passenger side being unreliable is clearly a quality/design issue.
Honestly, I actually love the Escape. The pedal feel is very responsive in all driving modes, compared in particular to the 2020 Hybrid Rav4, which felt like driving a boat (maybe I didn't find the drive mode?), or the 2020 VW Tiguan which had a shockingly slow automatic transmission for an ostensibly "sporty" vehicle. And I'm not even a car guy. I also love its actual buttons on the dashboard, instead of the idiotic "everything on a huge touchscreen" that too many cars do nowadays.
Cars here are inspected yearly anyway or you go change winter tires for summer tires. (Because we lack the place to store them in typical houses.) So you are at the garage anyway every 6 months to 12. Then they can also do the other stuff
(As a non American) I remember hearing a joke that goes something like “How do you fix a Chevrolette? Buy a Ford”, but nowadays I guess a bike is a better option
The new Tundra TTV6 had a manufacturing process defect that allowed shavings to get into the engine bearings, which causes catastrophic engine failure.
They still don't have a solution to the problem. The shavings amount/size is supposedly common among all engine manufacturing processes, but the new engine design has such tight tolerances that it's now problematic.
Actually in the latest J.D. Power initial quality ratings they took a big step up in quality. I think it was the first time in 15-20 years that they were on the list of recommended major brands.
The same Ford whose bean counters caused them decades of reputational damage over skimping on rust protection? Seems like they haven't learned any lessons at all.
I have spent a SOLID 3 full days 8h/day (plus long running tasks overnight) thrashing out a random idea for a Web application using purely Opus (mostly Max, sometimes ultracode version). I'm not a project manager, but I genuinely tried a full 3-tier spec out - design->specs->build details.
While it was significantly better than previous attempts, it still misses very basic things - sporadically. Eg. A clear design requirement was essentially adding clients, explained clearly and comprehensively. The ability to add clients was entirely missed in the build and iteration (there were multiple 'please check its all done' separate agent runs/checks).
I can imagine in a fully autonomous deployment, in even moderate complexity, even to this day would still occasionally mess up - badly enough to cause non-trivial business issues.
I haven't managed to really figure out what's the best way, but my latest thinking is really having boil down tasks to almost unit operations "add UI button, wire to Api call. End".
> there were multiple 'please check its all done' separate agent runs/checks
You could ask it to go through the spec point by point and then mark what is done and WHERE/WHY, then it'd point you towards exactly what might be missing.
Well, at least they learned from the experience, and that’s good.
The more interesting question, I think, is what proportion of businesses will choose the learn from Ford’s experience without first choosing to relive it?
Often people, and therefore also organisations, struggle to usefully learn from the experience of others without repeating the same mistakes, and experiencing the same pain.
The dystopian future where no one owns cars is already being laid.
Cars are more and more becoming white goods appliances with the driving experience becoming less and less a priority. Even enthusiast cars now are about raw numbers and need electronics to reign them in to make useable for the average driver on the average road.
The average user probably doesn’t even want to drive and have AI do it for them.
Repairability is becoming less viable as mechanical parts replaced with screens and digital locks. Parts availability is already an issue, only going to get worse especially with the pace of new cars are being churned out from China.
The end will be car as a subscription. We already have it with leasing, and BMW having to pay to use your electric seats.
> The dystopian future where no one owns cars is already being laid.
Pardon me?
We're living in the dystopian present, where most everyone has a car or several. Cities are crowded with cars -- both moving and parked -- and it's awful for humans who aren't cars.
I can't wait for the moment people switch to a subscription and the cars are shared and drive themselves. The streets will be just as full of moving cars, but at least the parked cars hopefully disappear, giving us more space for trees or sidewalks or anything but cars really.
You are injecting a lot of assumptions and wishful thinking to view the removal of ownership from this equation as a net positive.
I see no reason to assume that this would lead to the disappearance of parked cars or to more trees. Our corporate overlords will want to make use of that space for more cars or infrastructure to support the new car network, why would they ever just give it back willingly?
American tech is basically a sales machine. An ounce of tech will be coated with a ton of selling force. Everything in America is a business, presentation or a talk-show - including government, education, relationships. People do selling and faking to themselves sometimes.
> This has nothing to do with LLMs and instead is almost certainly about their MAIVIS and AiTriz pilots, which use old school CNNs on custom IBM hardware to do visual inspections.
At one point my boss asked why my AI usage was lower than other team members. I instantly knew what to do. Every session is now run at ultracode effort. My automated PR review bot averages like $80 in usage per PR review.
Oh and my favorite: Use 5 independent subagents to review code change and summarize the findings, and for any finding determine if they are real concerns
-Claude, burning my company's money.
Eg by doing that I was able to develop non-essential features which increased our quality of life for devs last month without going through our PO who'd need to price it - because that does let's you create changes in an incredibly hands off manner with miniscule amount of time investment if you already know what you want to achieve, and how the end result should be...
Admittedly, that's a pretty narrow usecase which is rarely the case- but if it is...
It's interesting that LLM barely had any vetting period or experimentation phase. Suddenly everyone was supposed to test it in production, it seems.
Get ready for that promotion!
There's so, so much mechanically simple but time consuming refactoring that should be done but nobody ever does that because there's never enough free time. Or even various utility scripts and at least finding out of date docs (or writing very basic ones where none exist, though it'd be hard to get them not to feel like slop writing). Or figuring out what additional custom linter rules would be useful, how to improve the CI pipelines and so on.
If I had the Anthropic Max 20x subscription, I could make a large part of the technical backlog disappear (relatively safely).
They can do what works, or they can fail. Large enough companies with enough inertia can do really dumb things for a while, but even giants fall.
Are you saying companies have to mandate AI everywhere?
Or are you saying the exact opposite, as your second sentence suggests?
I haven't heard of AI mandates in small companies, only in big ones.
That is essentially not an argument in any direction.
Quoting the host of the recurring Quiz Broadcast sketch from That Mitchell and Webb Look: "Books mention 'hope'. What was 'hope'?"
Value is measure in generalized labour, since that the universal measure of human effort. The genealized amount of time a human being must spend to produce something from its parts. Generalized labour is also what's bought from labourers. You don't pay them to do something specific, you pay them to labour in general.
This contrasts against specific labour, which is whats actually required in the moment. Generalized labour power must be the right kind of specific labour to actually produce anything of value.
The AI leaders have been told that AI is labour. To the extent that it currently is, which I believe is only the case because the market hasn't adjusted, it's not the right specific labour to male anything valuable.
You just have to get the input coefficient right. The least amount of acceptable quality with the least amount of costs is the sweet spot. /s
Ford has hired 350 engineers over the last 3 years which happened alongside short comings in using AI inspection tooling. This has nothing to do with LLMs and instead is almost certainly about their MAIVIS and AiTriz pilots, which use old school CNNs on custom IBM hardware to do visual inspections.
Dirt bag media will do anything for your clicks and leave you more uninformed at the other end.
After a quick search I found a publication actually mentioning about these tools:
Ford previously told Business Insider that it had developed two bespoke AI-enhanced scanning tools that helped validate that cars were properly assembled before rolling off the lot. The tools, called AiTriz and MAIVs, both debuted in 2024. https://autos.yahoo.com/policy-and-environment/articles/ford...
And after doing cursory research on these tools, it is clear they are rudimentary (as compared to SOTA LLMs), they were essentially smartphone mounted on stands and doing visual checks using the camera - so OP could be very right.
https://www.businessinsider.com/ford-uses-ai-cameras-in-fact...
(most of them are for fairly innocuous stuff...)
Few of the issues I've experienced with the car were clearly tied to quality issues: 1) Battery died a few times, but maybe that was user error 2) squirrels/rats nibbled the engine cable harness, a not-uncommon occurrence in our area. Only 3) auto-unlock on passenger side being unreliable is clearly a quality/design issue.
Honestly, I actually love the Escape. The pedal feel is very responsive in all driving modes, compared in particular to the 2020 Hybrid Rav4, which felt like driving a boat (maybe I didn't find the drive mode?), or the 2020 VW Tiguan which had a shockingly slow automatic transmission for an ostensibly "sporty" vehicle. And I'm not even a car guy. I also love its actual buttons on the dashboard, instead of the idiotic "everything on a huge touchscreen" that too many cars do nowadays.
The fact that you find this acceptable is amazing to me.
Sounds like a complete failure of quality control.
They still don't have a solution to the problem. The shavings amount/size is supposedly common among all engine manufacturing processes, but the new engine design has such tight tolerances that it's now problematic.
None of the US automakers have good quality reputations. If you want something that works reliably, get a Toyota.
https://archive.is/VcL8c
While it was significantly better than previous attempts, it still misses very basic things - sporadically. Eg. A clear design requirement was essentially adding clients, explained clearly and comprehensively. The ability to add clients was entirely missed in the build and iteration (there were multiple 'please check its all done' separate agent runs/checks).
I can imagine in a fully autonomous deployment, in even moderate complexity, even to this day would still occasionally mess up - badly enough to cause non-trivial business issues.
I haven't managed to really figure out what's the best way, but my latest thinking is really having boil down tasks to almost unit operations "add UI button, wire to Api call. End".
You could ask it to go through the spec point by point and then mark what is done and WHERE/WHY, then it'd point you towards exactly what might be missing.
The more interesting question, I think, is what proportion of businesses will choose the learn from Ford’s experience without first choosing to relive it?
Often people, and therefore also organisations, struggle to usefully learn from the experience of others without repeating the same mistakes, and experiencing the same pain.
Cars are more and more becoming white goods appliances with the driving experience becoming less and less a priority. Even enthusiast cars now are about raw numbers and need electronics to reign them in to make useable for the average driver on the average road.
The average user probably doesn’t even want to drive and have AI do it for them.
Repairability is becoming less viable as mechanical parts replaced with screens and digital locks. Parts availability is already an issue, only going to get worse especially with the pace of new cars are being churned out from China.
The end will be car as a subscription. We already have it with leasing, and BMW having to pay to use your electric seats.
Pardon me?
We're living in the dystopian present, where most everyone has a car or several. Cities are crowded with cars -- both moving and parked -- and it's awful for humans who aren't cars.
I can't wait for the moment people switch to a subscription and the cars are shared and drive themselves. The streets will be just as full of moving cars, but at least the parked cars hopefully disappear, giving us more space for trees or sidewalks or anything but cars really.
I see no reason to assume that this would lead to the disappearance of parked cars or to more trees. Our corporate overlords will want to make use of that space for more cars or infrastructure to support the new car network, why would they ever just give it back willingly?
the ~game~ matrix
Our AI sucked but that doesn't mean less AI. We need better AI, not humans.
Buy a BYD / Xiaomi / Zeekr / Xpeng...
https://news.ycombinator.com/item?id=48704222
Reminds me of this disaster at Toyota,
https://www.wsj.com/business/autos/toyota-bet-technology-wov...