I see things like 2 sentence menu summaries in Uber Eats that are completely off in tone.
A quick sample from my app right now:
“Authentic Caribbean Flavours. Jerk Chicken, Curry Goat, and more. A vibrant culinary journey awaits.” - local Caribbean place
“Customisable burgers with 250,000+ toppings. Hand-cut fries and rich milkshakes await.” - Five Guys
“Authentic Indian cuisine bursting with rich flavours. Perfect for late-night cravings” - local Indian
Everything is Authentic, or Rich, or whatever.
—-
They’re investing in the wrong bits of AI. I’m sure they’re AB testing these soulless often inaccurate blurbs but I just cannot see how investing money into them actually sells more product.
On the other hand, if they had a coherent product vision, and trusted their engineers to use AI how they see fit, then I’m sure they would be more successful, and it would be cheaper.
Aside from the hilarious "250,000+ toppings" error, these summaries seem... fine? I would be unsurprised to learn that a human came up with them, even. Seems like pretty common/standard marketing copy.
Maybe each one is fine in isolation - what doesn’t come across from the sample is that every single one is practically the same. If you have Uber Eats, open up the app and look through the summaries for a bunch of restaurants and you’ll see what I mean.
And besides that, this just feels like something nobody asked for that probably doesn’t sell more food compared to, for example, more pictures.
Don't worry, they'll also use AI to add more pictures, which will all look strangely similar while bearing at best passing resemblance to anything you might actually receive after placing an order.
I think that's exactly the point. It's the distillation of the most common marketing copy possible, and when that tone is applied everywhere it becomes very same-y, like those cookie cutter neighborhoods where every house is the exact same. Which to some extent defeats the purpose of marketing as it doesn't stand out at all, just sanitized sameness. It's boring and a bit creepy.
But why does Uber need to spend 3.4B on injecting a useless blob of text between me and an overpriced burger delivered by a struggling illegal immigrant in a smoke-belching jalopy?
I know the counter-argument. "This will increase sales". You know what else would increase sales? Spending the 3.4B to replace the above with a uniformed delivery service similar to UPS. That job could pay benefits.
> if they had a coherent product vision, and trusted their engineers to use AI how they see fit, then I’m sure they would be more successful
Out of curiosity, what do you think might be a successful application for AI in Uber's business? It seems like this is the sort of thing AI applications end up being. Does it actually get better than this?
> According to The Information, Chief Technology Officer Praveen Neppalli Naga said Uber is now "back to the drawing board" after a surge in the use of AI coding tools, particularly Anthropic's Claude Code, has blown past internal expectations.
Of usage costs?
> The payoff is starting to show. Around 11% of Uber's live backend code updates are now written by AI agents, up sharply in just a few months. These systems power everything from ride-matching to pricing and bug fixes.
That's not a payoff.
What is the immediate cost of those code updates, what is the quality, how do they affect longer-term maintenance, how does that compare to doing it without "AI", etc.
Are these articles written to inform or to hype?
> UNLOCKED: 5 NEW TRADES EVERY WEEK. Click now to get top trade ideas daily, plus unlimited access to cutting-edge tools and strategies to gain an edge in the markets.
There's my answer. Here's a helpful uBlock Origin filter:
Yes, yahoo “journalism” is garbage. The primary source of this story is paywalled, so I can’t actually see what it said, but this AI (or otherwise crappy) summary is worthless.
>Despite spending $3.4 billion on research and development, the company has already exhausted its planned AI budget just months into 2026.
This, and the rest of the article, does does not seem to support that they spent 3.4B on AI. The text implies that the R&D budget for the entire company is 3.4 billion (which sounds vaguely reasonable given that market cap), and the portion of that which was earmarked for AI is already spent. I have no idea what the AI spend is there (although I assume it's not small), and the article doesn't provide any number either.
Those are extremely different things (unless there's evidence that 100% of R&D is spent on AI) and that headline seems to be intentionally misleading.
I'm coming around to it being like getting a pair of industrial grade yak clippers. Yes, there will be a lot of shiny yaks, but the market for shiny yaks is low.
I'm a hot dog chef with over 20 years of experience. Credited with inventing 274 hot dog styles. International awards. World renowned and industry figure.
My entire team, very competent hot dog experts, was laid off after a hot dog cooking machine could do what took us 3 months, in just one day. I've been out of a job for 12 months. The reason? All hot dog making has been offloaded to Claudog Hotdog. "Sorry. Hot dog manual cooking is a thing of the past", one recruiter told me.
I'm working as a software engineer as we speak. I keep applying to hot dog related positions but I get no interviews. Even positions significantly below my pay grade and skillset. No one is hiring. Hot dog cooking is over. We are entering a new era.
I'd take these options from several companies (all selling hotdogs) and wrap them up in Collateral Hotdog Obligations which I'd then offer to investors.
The main question is: what is demand elasticity for software?
If it low, and lower prices won’t generate much new demand, we should expect AI to improve engineering productivity, and for companies to reduce staff.
If it is high, then we should see companies hire more engineers, increase output and lower prices (and earn more).
Bureaucracy creates work so long as it owns the production function. In software that's typically through system upgrades, new API's, etc. The system will grow in internal complexity to its carrying capacity. You'd need someone who understands how to replace parts to prune, but they don't really have the incentive. This effect is reduced where software is less essential to the product, but any software-heavy product (particularly with a moat) will be more susceptible.
Companies try to manage it via CI/CD, outsourcing and internal competition, but no, companies can't magically reduce staff. They can, however, inject fear, which is good for reducing overt bureaucratic games, but actually increases covert bureaucracy and reduces knowledge-sharing, making the problem worse.
Only when incentives are aligned - when developers have an (equity) stake in growing the company - can the culture be open and efficient.
Every time the cost of software development has gone down due to higher level tools we've gotten higher software budgets, more software developers, and more released software. Demand appears to be effectively infinite.
oh man uber is acquiring the company I work for [1] and we currently really like Claude ... but if Codex is better so be it. I just really, really, really like Claude Code as a front end. Guess I'll have to make it talk Codex instead.
If it is anything like my company, sign enormous deals to AI startups that have existed for 8 months, and do little more than provider wrappers around someone else's model. Then hire three different firms that do the same thing because each division has to prove how much more AI they are than the others. Have a handful of internal engineers who have no idea what they are doing, but get approval to build and run an internal B200 server farm. Ensure any big jobs are done through some kind of white-glove offering from Amazon/Azure that removes complexity, but charges astronomical rates.
This makes it sound like they spent $3.4B on tooling, but is it actually on salaries? Hardware?
Probably 5k-6k hires in the department, at say $350k/employee costs, is $2.1B which still leaves a ton of extra costs somewhere. Are they sending $1B to Anthropic?
Large companies have been incentivizing and correlating token spend to performance, thus creating needless spend of tokens for now. Goodharts Law and all that.
Holy misleading headlines Batman. They're not spending $3.4B on solely tokens for Anthropic are they? I don't think so...
If anything the CTO is just saying, we're blowing through token budgets way faster than expected as the uptake is so immense. I think that's right from what I've seen. Once people get it, they start using AI for everything. Obviously that's not going to be sustainable forever. I do think we're going to see a lot of adaptive routing in the future to cheaper models for more mundane tasks, whereas right now everything is getting routed to Opus regardless of real need.
Why would that not be sustainable forever? Over the long run the price per token is likely to decline as both hardware and software gets more efficient.
Look, we're using Uber Eats to order food for "free food Tuesdays" in our office.
I'm struggling to not puke using their interface, and a couple of times I gave up ordering even though it was free.
Every click can take 2-5 seconds to be processed, without any indication. Menus glitch. I once got 2 copies of my order because I rage-clicked the "Finish" button several times.
So you're trying to do high-end AI when you can't make a basic fucking form-based webapp work?!? What do you expect?
I see things like 2 sentence menu summaries in Uber Eats that are completely off in tone.
A quick sample from my app right now:
“Authentic Caribbean Flavours. Jerk Chicken, Curry Goat, and more. A vibrant culinary journey awaits.” - local Caribbean place
“Customisable burgers with 250,000+ toppings. Hand-cut fries and rich milkshakes await.” - Five Guys
“Authentic Indian cuisine bursting with rich flavours. Perfect for late-night cravings” - local Indian
Everything is Authentic, or Rich, or whatever.
—-
They’re investing in the wrong bits of AI. I’m sure they’re AB testing these soulless often inaccurate blurbs but I just cannot see how investing money into them actually sells more product.
On the other hand, if they had a coherent product vision, and trusted their engineers to use AI how they see fit, then I’m sure they would be more successful, and it would be cheaper.
Aside from the hilarious "250,000+ toppings" error, these summaries seem... fine? I would be unsurprised to learn that a human came up with them, even. Seems like pretty common/standard marketing copy.
Maybe each one is fine in isolation - what doesn’t come across from the sample is that every single one is practically the same. If you have Uber Eats, open up the app and look through the summaries for a bunch of restaurants and you’ll see what I mean.
And besides that, this just feels like something nobody asked for that probably doesn’t sell more food compared to, for example, more pictures.
Don't worry, they'll also use AI to add more pictures, which will all look strangely similar while bearing at best passing resemblance to anything you might actually receive after placing an order.
I think that's exactly the point. It's the distillation of the most common marketing copy possible, and when that tone is applied everywhere it becomes very same-y, like those cookie cutter neighborhoods where every house is the exact same. Which to some extent defeats the purpose of marketing as it doesn't stand out at all, just sanitized sameness. It's boring and a bit creepy.
But why does Uber need to spend 3.4B on injecting a useless blob of text between me and an overpriced burger delivered by a struggling illegal immigrant in a smoke-belching jalopy?
I know the counter-argument. "This will increase sales". You know what else would increase sales? Spending the 3.4B to replace the above with a uniformed delivery service similar to UPS. That job could pay benefits.
That's not an error, it's what 5 Guys advertise. It's the number of combinations for their toppings.
That sounds like an error to me. "Number of toppings" is not the same as "Number of possible combinations of toppings".
Shouldn’t it be a lot more? Around 20 toppings in any combination and count would be 20! + 19! + 18! … no?
2^20
Does order of toppings matter?
yes, only a monster would put the pickles underneath the tomato
The article seems to suggest the unexpected spend was primarily on coding tools, like Claude Code.
One would hope Uber could manage 1 sentence API summaries (regardless of their quality) for less than $3.4 billion.
I've never cared about those menu summaries! I always look at menu items and their descriptions. They are fine, at least to me.
More than half the menu item descriptions I see now are AI generated, and some are completely inaccurate.
You're absolutely right.
I am disappointed that not all your comments are this line :D
i have a solid picture now, the honest answer is: you're absolutely right
That's a very insightful observation, highlighting the genuine tension between consistent messaging and quick, pithy responses on Hacker News.
> if they had a coherent product vision, and trusted their engineers to use AI how they see fit, then I’m sure they would be more successful
Out of curiosity, what do you think might be a successful application for AI in Uber's business? It seems like this is the sort of thing AI applications end up being. Does it actually get better than this?
The only accurate summary is "shit and overpriced"
> According to The Information, Chief Technology Officer Praveen Neppalli Naga said Uber is now "back to the drawing board" after a surge in the use of AI coding tools, particularly Anthropic's Claude Code, has blown past internal expectations.
Of usage costs?
> The payoff is starting to show. Around 11% of Uber's live backend code updates are now written by AI agents, up sharply in just a few months. These systems power everything from ride-matching to pricing and bug fixes.
That's not a payoff.
What is the immediate cost of those code updates, what is the quality, how do they affect longer-term maintenance, how does that compare to doing it without "AI", etc.
Are these articles written to inform or to hype?
> UNLOCKED: 5 NEW TRADES EVERY WEEK. Click now to get top trade ideas daily, plus unlimited access to cutting-edge tools and strategies to gain an edge in the markets.
There's my answer. Here's a helpful uBlock Origin filter:
Yes, yahoo “journalism” is garbage. The primary source of this story is paywalled, so I can’t actually see what it said, but this AI (or otherwise crappy) summary is worthless.
>Despite spending $3.4 billion on research and development, the company has already exhausted its planned AI budget just months into 2026.
This, and the rest of the article, does does not seem to support that they spent 3.4B on AI. The text implies that the R&D budget for the entire company is 3.4 billion (which sounds vaguely reasonable given that market cap), and the portion of that which was earmarked for AI is already spent. I have no idea what the AI spend is there (although I assume it's not small), and the article doesn't provide any number either.
Those are extremely different things (unless there's evidence that 100% of R&D is spent on AI) and that headline seems to be intentionally misleading.
”Engineers were actively encouraged to use tools like Claude Code and Cursor, even ranking them on internal leaderboards based on usage”
Token maxxing? Might explain high costs if you are actively encouraging developers to spend as much tokens as possible.
I was told AI makes people more productive so the costs should easily pay for themselves in the form of more revenue.
I'm coming around to it being like getting a pair of industrial grade yak clippers. Yes, there will be a lot of shiny yaks, but the market for shiny yaks is low.
I’d be first in line for a three piece qiviut suit!
Maybe the lesson here is we shouldn't rely on the guys selling picks and shovels when determining if we should be buying picks and shovels.
I’m the CEO of a hot dog company. I’ve worked on hot dogs for 10 years. And I wasn’t prepared for what I’ve just seen. Your life is about to change.
So what can you do?
Buy as many hot dogs as you can. Buy stock in hot dog companies.
I'm a hot dog chef with over 20 years of experience. Credited with inventing 274 hot dog styles. International awards. World renowned and industry figure.
My entire team, very competent hot dog experts, was laid off after a hot dog cooking machine could do what took us 3 months, in just one day. I've been out of a job for 12 months. The reason? All hot dog making has been offloaded to Claudog Hotdog. "Sorry. Hot dog manual cooking is a thing of the past", one recruiter told me.
I'm working as a software engineer as we speak. I keep applying to hot dog related positions but I get no interviews. Even positions significantly below my pay grade and skillset. No one is hiring. Hot dog cooking is over. We are entering a new era.
You just need to pick yourself up by the bootstraps and start your own artisanal hand-grilled hotdog food truck.
> I'm working as a software engineer as we speak.
Well, there’s your problem. Get back to making the hotdogs!
I will gladly pay you Tuesday for a hotdog today (if you want to get a circular investment bubble going)
Do you sell hotdog options? I want some hotdogs at a guaranteed price, but only next week. (I just want upside exposure)
I'd take these options from several companies (all selling hotdogs) and wrap them up in Collateral Hotdog Obligations which I'd then offer to investors.
Collateralized hotDog Obligations made up of MBS (Mostly Bones and Sawdust) would be a financial product I could sell to institutional investors!
I will sell you hotdog default swaps. Even if I lose, I win.
Is a synthetic hotdog default swap considered vegan in Cali?
I suppose a discount code is effectively a hotdog option.
The main question is: what is demand elasticity for software?
If it low, and lower prices won’t generate much new demand, we should expect AI to improve engineering productivity, and for companies to reduce staff.
If it is high, then we should see companies hire more engineers, increase output and lower prices (and earn more).
Bureaucracy creates work so long as it owns the production function. In software that's typically through system upgrades, new API's, etc. The system will grow in internal complexity to its carrying capacity. You'd need someone who understands how to replace parts to prune, but they don't really have the incentive. This effect is reduced where software is less essential to the product, but any software-heavy product (particularly with a moat) will be more susceptible.
Companies try to manage it via CI/CD, outsourcing and internal competition, but no, companies can't magically reduce staff. They can, however, inject fear, which is good for reducing overt bureaucratic games, but actually increases covert bureaucracy and reduces knowledge-sharing, making the problem worse.
Only when incentives are aligned - when developers have an (equity) stake in growing the company - can the culture be open and efficient.
Every time the cost of software development has gone down due to higher level tools we've gotten higher software budgets, more software developers, and more released software. Demand appears to be effectively infinite.
>what is demand elasticity for software?
NP-hard
3.4B in 4.5 months…is that all going to Anthropic? Makes it seem so with the wording and how they’re pivoting to Codex too
It's probably all AI spending, including them doing AI stuff for their products.
oh man uber is acquiring the company I work for [1] and we currently really like Claude ... but if Codex is better so be it. I just really, really, really like Claude Code as a front end. Guess I'll have to make it talk Codex instead.
[1] it's public knowledge https://investor.uber.com/news-events/news/press-release-det...
Curious how it works in other countries, do employees get a portion of the payout?
If it is anything like my company, sign enormous deals to AI startups that have existed for 8 months, and do little more than provider wrappers around someone else's model. Then hire three different firms that do the same thing because each division has to prove how much more AI they are than the others. Have a handful of internal engineers who have no idea what they are doing, but get approval to build and run an internal B200 server farm. Ensure any big jobs are done through some kind of white-glove offering from Amazon/Azure that removes complexity, but charges astronomical rates.
Apparent source: https://www.theinformation.com/newsletters/applied-ai/uber-c...
This makes it sound like they spent $3.4B on tooling, but is it actually on salaries? Hardware?
Probably 5k-6k hires in the department, at say $350k/employee costs, is $2.1B which still leaves a ton of extra costs somewhere. Are they sending $1B to Anthropic?
Weird and uninformetive article.
Large companies have been incentivizing and correlating token spend to performance, thus creating needless spend of tokens for now. Goodharts Law and all that.
Holy misleading headlines Batman. They're not spending $3.4B on solely tokens for Anthropic are they? I don't think so...
If anything the CTO is just saying, we're blowing through token budgets way faster than expected as the uptake is so immense. I think that's right from what I've seen. Once people get it, they start using AI for everything. Obviously that's not going to be sustainable forever. I do think we're going to see a lot of adaptive routing in the future to cheaper models for more mundane tasks, whereas right now everything is getting routed to Opus regardless of real need.
Why would that not be sustainable forever? Over the long run the price per token is likely to decline as both hardware and software gets more efficient.
Look, we're using Uber Eats to order food for "free food Tuesdays" in our office.
I'm struggling to not puke using their interface, and a couple of times I gave up ordering even though it was free.
Every click can take 2-5 seconds to be processed, without any indication. Menus glitch. I once got 2 copies of my order because I rage-clicked the "Finish" button several times.
So you're trying to do high-end AI when you can't make a basic fucking form-based webapp work?!? What do you expect?