I never took tokenmaxxing to be about improving productivity directly; mundane feature work that comes out of it is just a side effect. I always saw it as a race between these big tech companies to get a generational advantage by being the one to discover the way of the future, with respect to harnessing AI to actually and truly automate software development.
EDIT: whoa, I used "way of the future" as a reference to Howard Hughes in "The Aviator", not this Way of the Future religious organization thing I just stumbled on; no intended reference there.
My impression of companies pushing AI so heavily is that they are basically being forced to do it by it merely existing. Imagine if AI really is as powerful as it is suggested to be and you didn't jump on the bandwagon. Then you would be behind. So by it existing and other companies using it, you have to as well because even if it turns out to be a failure at least everyone else will have failed too and you are on an even playing field.
I feel like this is kind of like the gambler's risk of "going bust" but applied backwards?
Feels like it would be better to spend just enough so that you have the capacity to scale up IFF LLM's end up being a big deal. You spent less than your rivals (who were competing for a supremacy that never came), but you have saved more "dry powder" to compete against them for the more likely future. The only future you exclude with this strategy is the "LLM Supremacy" future, which you only had a 1/(number_of_players) chance of winning anyway. :]
I think the real reason for the spending is that the scent of LLMs in the air causes stock values to go up. And even if everyone knows it does not make sense, they still want "NUMBER GO UP", and so they will spend more money to excite the amateur and professional investor class.
There is a fad every few years that uses similar thinking to justify spending. Sometimes they work out sometimes they don't. But they never work out as good as the most optimistic predictions.
I commented on wasteful AI spending, and my father immediately started talking about how he "went through that a couple of times" as a manager at an oil company.
The unusual thing is perhaps how global and cross industry it seems.
Genuinely asking: for which fads was it actually beneficial to jump in during the hype phase? Was there ever anything so critical that there was some huge disadvantage if you didn't adopt it right away?
ETA: I suppose the complicating factor, at least for B2B, is "customers demanding $fad", particularly when the purchasing decision makers don't actually understand what $fad is (e.g., "cloud", "blockchain", "ai", ...). If you don't become "$fad native" right away, you lose the Dunning-Kruger segment of the market.
That's what throws me off about the AI hype. I've lived through plenty of other hype cycles, but none had such rapid adoption as this during the hype phase. Usually its a bunch of early adopters and a few doomed startups, with the big established (non tech) companies never fully jumping in, and then by the time they might have considered it, the fad died.
Even "cloud" which did stick around and actually did pan out, didn't see such immediate adoption during the hype. There were a lot of companies that stayed on-prem for a long time, many which still are, and none of them imploded for not jumping on the hype.
Why is the FOMO so strong with AI this time around? I don't ever recall being told "spend as much money on AWS as you possibly can!" during the cloud hype...
because the tech industry is at a hole is out of creative ideas. That why ever seen so many hype cycles since the end of zirp - and why the only innovation we're seeing is on ways to squeeze more money out of people. and AI was sold as a salary cutting magical money machine... AKA one more hype to jump on.
except this one a isn't making anyone rich besides Nvidia
> because the tech industry is at a whole is out of creative ideas
This isn't true. However the tech industry is out ideas that apply to many many people and scale well.
Most people need a word processor at some point in their life (if your school doesn't make you write at least one paper on a word processor then your education failed - you might never do it again in your life but it is still an important skill), but those were already powerful enough in the 1980s, and the 1990s solved most of the usability issues.
Robotic vacuums can get some more innovation, but the obvious next steps are unsolved problems (I want it to pick up before it cleans) that may not be solvable for a reasonable price.
There are however a ton of niches that could use more technology. However because they are niches they don't scale. You can make millions (gross profit) if you can solve their problems, but not tens of millions - this is enough to get your personally a nice lifestyle if you run or work for such a company (think a 3-5 person company), but even if you could interest an investor there isn't enough for them to skim off any profit and still make money.
Not all niches that need tech are that small. There are a few large ones, but they are hard to find (if they were easy someone would have done it already). There are also a lot of what looks like large ones that either are not large, or are not large enough to pay for the investment needed. There are also some medium sized places that tech can help. Once in a while there are even tiny places where someone can make a difference (but generally this means you do the thing as your business and tech as a hobby after work)
> Robotic vacuums can get some more innovation, but the obvious next steps are unsolved problems (I want it to pick up before it cleans) that may not be solvable for a reasonable price.
Roborock has already released a vacuum that does this. From what I've seen it's limited, but it seems to work for the things it can pick up.
Sears, the obvious choice, could have been Amazon. There are plenty of other less extreme examples of defunct brick-and-mortar stores who were killed by internet competitors (often, Amazon).
Yeah this is right. Its about taking the throttle off of experimentation hoping some team suddenly starts shipping a years worth of features in a week and responding to strategic customer demands in near real time. Then copying what works out across the org. (and probably downsizing)
What those organizations miss is that they drown their teams in organizational red tape. The way to the future isn't tokenmaxxing - it's cutting back on process noise.
It was very easy to see before this started that the models were just repetitive inflators.
It’s not that they should “scale back” their use as much as the metric should be improvement/tokens. Tokens used is a denominator in any worth calculation.
> by being the one to discover the way of the future
This is my understanding too. The underlying assumption is that action leads to information, iterations lead to enlightenment. So from an org's point of view, tokenmaxxing means encouraging everyone to explore as much as they can. Of course, token volume should not be the only metric - tokenmaxxing is just a catchy phrase.
> action leads to information, iterations lead to enlightenment
So doing something (action) creates something new (more information), and iterating on that new information leads to the realization there is nothing new left to be learned with that information (enlightenment). Is how I'm interpreting that.
To the extent that figuring out the true automation will lead to layoffs, that knowledge will quickly spread as the laid-off knowledge workers are available for hiring.
Like many corporate behaviors I believe we tend to rationalize them too much. Most job posts for management ask for previous experience with "implementing AI". Managers don't want to be left out any more than devs. Devs rewrite everything in the language-du-jour, managers max out tokens.
I agree.
We don't need a nondeterministic 10quadrillion vector model.
We need an deterministic expert on our narrow business.
Something small, that can be run on the 2026 version of the spare PC under the CTOs desk.
It will always be worse compared to a centralized approach where hardware utilization can remain high. Except in case which demand low latency which most development things do not need. It's okay if it takes an extra 100ms for a code review to take place.
It's little different from the mainframe/mini computer to the PC: Huge servers and resources will be better--no dispute there--but good-enough will be achieved using local resources.
There are privacy and general de-centralization reasons to prefer this outcome, even though most AI and cloud-first tech companies don't want this.
How long it will take us to get this point is a different matter.
By that logic shouldn’t streaming video games also be centralized?
Yet startups keep trying it and failing. Turns out users actually want exclusive access to that hardware to have a smooth experience. The tradeoff has always been between faster exclusive hardware or slower but cheaper shared hardware.
If local hardware can’t beat shared hardware on performance then something’s wrong? Either it’s because the providers are charging wildly below cost or because local hardware just hasn’t needed to catch up. Maybe it’s both.
Agree and I have wondered behind close doors if this is not the mental model. You need to spend money to see what is working this was simply a way to see that.
Eh, I also saw it as a rather blatant attempt to undermine the bargaining power of the Software Engineer (which had grown to insane proportions over the years) - both in working conditions and raw cash money.
In many cases it really didn’t/doesn’t matter if the AI automation actually works, just that people think it could - and hence leave money on the table.
Now feels like a very good time to be a small team of experienced developers who can largely work on stuff by themselves and not a corporation of hundreds of developers of varying abilities all now trying to show how much code they can generate and how many tokens they can burn.
100% - I think that is also part of the divide you see online. Devs who work on massive codebases w/ 100s of engineers and see the bugs the LLMs create vs devs who work on smaller codebase w/ small <5 person team.
Small teams building continuously get to write features that are 90% correct in a tenth of the time.
Big enterprises get to write features that are 90% correct barely twice as fast, because all of the bottleneck lies elsewhere. They also spend more on AI per user because of the internal dynamics pushing people to adopt AI irresponsibly. They can correct the 10% of errors slower than small teams because of bureaucracy, increasing the cost of errors that show up in the product. Furthermore, they have less to gain from a given amount of speedup because they had plenty of engineering velocity anyway compared to small teams.
I don't think big enterprises will start winning from AI technology until AI truly can automate almost everything in a company and let said company outproduce competitors by burning tokens alone. That's nowhere near possible right now.
I don't think "90% correct in a tenth the time" is really the tradeoff. For a well-specified task, it's closer to 100% correct in 1/100th the time. But that undersells the time required to generate a good spec.
For under-specified tasks, it's not really accurate to talk about "correctness," because the machine isn't psychic. I would suggest that given a high-level feature request like "add streaming support" it's more about acceptance probability. In a well-structured and well-documented codebase, and a reasonably sized feature request, there might be an 80% chance it will generate something which is 100% acceptable. But there's about a 99% chance it will generate something which is acceptable after 1-2 revisions.
Even small teams are not immune from the AI push. I work in a small <10 people with 3 devs. I'm the only one not using AI while receiving push that I'm a problem for not. To the point, VP of company says if we don't all start using AI we'll be out of business because other people will. :face-palm:
I don't like outsourcing learning and prefer to do the actual learning. Just like taking notes via recorder vs writing them down during a lecture causes you to not remember the information as well. Allowing something else to create your code tends to mean you're not going to remember/know the how/why it works and you just accept it works and moves on. Eventually you don't know how any of it works and you're totally dead in the water without the LLM. My understanding of the code I write is much more valuable than some arbitrary deadline that means nothing
Do you also inspect and study what assembly code your program was compiled to?
// Obviously LLMs are non-determenistic etc and it depends on your domain, but your VP's point 100% makes sense if you folks are trying to cook up another demo-CRUD apps to convince investors for another funding round
I keep hearing this argument on HN. Yes, if performance is something you care about you totally look at the disassembly. You don’t even need to write assembly, just be able to read it. By now I assume that programmers who argue they don’t need to understand and review what the LLM generates for them are simply not that skilled and don’t realize that skilled programmers do this.
If you're just building a non-security-sensitive frontend to validate market traction, it makes total sense to go full AI. The commenter working at a 10-person company getting flamed for not using AI to iterate aggresively against competitors has a super valid point.
Validation methods will evolve to accommodate human laziness. Insisting on doing it the hard way is no different than the old-timers who used to claim engineers 'weren't skilled' if they didn't know how to use punch cards.
Ad hominem. Skilled programmers != old-timers calling others unskilled. Who does fuzz testing, automated integration tests, etc already, that LLMs need as a test harness? It’s not the vibe coders saying “nobody reads assembly anyway” as an argument.
I'd definitely say IntelliJ improved my productivity. No one cared. If anything, management viewed it as a nuisance as they had to deal with the license.
Now there is demands to justify not using AI like this, but people don't care about details. Which AI tool I use apparently doesn't matter at all, even if there are presumably productivity differences between them.
Seems like you should have a justification for introducing a new tool instead, no? There are thousands of other tools they're also not using, but we're not asking about those.
I'm not the same person, but I don't really use it. I write better code than an LLM, and I do it faster than the time it takes to review and correct the LLM's bad code. The tool has no value to offer me.
I think the jury’s still out on how big an impact AI will have on overall, average productivity. But it’s definitely a productivity boost for someone who’s capable of writing code without it. If you want to be super conservative, don’t even have it write any code. Use it to search through existing codebases, review your code, find the root cause of bug reports, evaluate pros and cons of alternative approaches, etc. etc. You’re really missing out by not using it at all.
Here’s a concrete example of conservative AI usage: I use Claude to vibe code my nvim config. Now, who cares if my nvim config is AI slop? What’s the worst that can happen? Nvim works for me now way better than it ever did when I was limited by the time I was willing to spend configuring it manually.
A friend of mine added some pretty extensive iOS UI tests to a keystone feature hit by millions every month. They'd been kicking the can down the road for years, trying to fit it in their roadmap, and with Claude running overnight they were able to bang out the whole suite in a week.
I'm not sure how it would show up in quarterly results.
I see these kinds of stories here a lot, and I'm curious whether they reflect a steady stream of need for AI coding, or whether a lot of companies have a burst of AI-appropriate coding work now that the technology is available and then will have a smaller need going forward.
Is it like the stereotypical dad who rents a power washer, powerwashes every exposed surface on his property, and then doesn't need to do any powerwashing for a few years; his neighbor who gets an Instant Pot and uses it for every meal for a month, then sees it gathering dust when the family gets tired of pressure-cooked stews; or like their neighbor who gets a microwave oven and uses it multiple times a day for decades?
So far where I work its the Instant Pot, at least for the non-devs. We rolled out Claude & Cowork to the masses after a brief pilot. It was about a solid month and a half of heavy usage and then suddenly usage fell off a cliff. Once it stopped being a cool new toy, people just didn't find a use for it.
A few mundane things got automated, but these were just back office admin type work. Nothing that's going to show on the P&L. Yeah those people now have a little more time for other things, but those other things are also not revenue generating. No FTE got replaced by it so in the end they just paid for a bunch of administrative positions to be a little less busy. Great for the workers who are now less stressed, but almost no impact on the business financials except there's now yet another subscription.
> So far where I work its the Instant Pot, at least for the non-devs. We rolled out Claude & Cowork to the masses after a brief pilot. It was about a solid month and a half of heavy usage and then suddenly usage fell off a cliff. Once it stopped being a cool new toy, people just didn't find a use for it.
Your employer is doing it wrong. You need usage surveillance with sanctions for low/declining use, then people won't stop using it.
It's industry best practice. All the market-leading companies are doing it. Do you think you know better than them?
If there's anything I've learned as a software engineer, it's that agreeing with and defending the ideas of business leaders and Silicon Valley VC influencers proves I'm very intelligent.
That’s been my theory - there’s some low hanging fruit in every environment where AI knocks it out of the park. Then complex brownfield reality (coupled with non-technical factors) rears its head and the stunning productivity gains are nowhere near to be seen.
That’s the explanation how you can have both the anecdotes of amazing AI productivity and rigorous studies showing anything from actual loss of productivity to single-digit gains.
I also think in addition to that the increased speed compounds the problems much quicker. And I don't mean bugs. I mean that duplicated code here, that additional state variable to keep everything going there. Not removing things that should be removed because we can work around that, etc.
It's like building a super tall Jenga tower very quickly but laying the bricks much worse than a careful player.
I've found it to be that people simply go back to their old ways. Many people are busy and unless they're being actively pushed to change it they feel they only have time to do things the way they have been doing them.
Using AI requires skills to know how to use it, particularly agents then it requires the time to build an improved way of doing things.
I think of it like giving someone excel in the beginning and expecting they know how to use it, when the rest of their team doesn't, they don't have the skills to know how to use it and how it can benefit them.
> or whether a lot of companies have a burst of AI-appropriate coding work now that the technology is available and then will have a smaller need going forward
For the product my friend works on, it's definitely the latter. I definitely don't expect this party to last forever.
Nah in competitive industries you need to build features and out compete people and getting AI to do that whilst architecting things well due to experience and having time to think more about the important stuff but have a lot of the more boilerplate and simple things ABs plumbing etc handled by agents is great.
When you try to replace your entire brain with AI things are going to go wrong.
Some measures should have real, tangible, concrete numbers; others should have “my friends are saying”/“you are blind if you are not seeing it” vibing.
Everything can be expressed in financial terms- it is one of the guiding principles of the universe. Anyone who thinks they are above or beyond it will rue the day.
Yes but that representation can be over-applied and misleading. For example you can roughly estimate the price of the entire earth. What does that number mean though? Well, nothing, really.
In my limited experience with using agents to create tests it tends to code the tests to the existing code instead of ensuring the correctness from a spec. Great for regression testing but still limited in effectiveness for catching existing issues.
Over here our CTO replaced Intercom with an internal equivalent that costs less than $20 / month to run, haiku and sonnet support agent costs included. In less than a few weeks, in his spare time.
Even if Uber really did double developer productivity, would it translate to quarterly results?
Ultimately they make money selling rides, not selling software. The Uber app is mature and adding new features is unlikely to significantly increase sales.
Writing 2x more code doesn't translate to 2x more revenue unless it results in 2x more rides.
How much is infra spending compared to revenue/other expenses? I honestly can’t imagine it’s that high, they’re not running something like Netflix or YouTube… But maybe I’m underestimating it.
Anthropic's annualized run rate is >$40b according to outside reporting. AWS hit that by Q4 2019. There were still debates on public cloud vs on prem at that time, but by late 2019 public cloud had facilitated the creation or adoption of entire categories of software within SaaS and PaaS, not to mention consumer internet businesses like Uber and Airbnb. The net impact of AI coding tools is far more ambiguous in comparison.
The profitability comparison is fraught but worth noting that by then AWS was already extremely profitable.
Amazon was AWS first customer. I sometimes feel if the AI companies promises of their models replacing all SWE, they would be a top 3 product/service in any number of businesses and not selling shovels to miners.
But they do use their models internally quite a bit. Claude code head talks about how he doesn't write any code himself anymore and just uses agents to do everything. So they are using it themselves.
I actually think Chinese models have already popped the bubble, we just dont see it yet. the only way to justify AI IPO market cap is basically if they get to hold most sw industry code hostage and then token-flate budgets to collect AI tax. short of that AI expense would very quickly mean revert to model + some margin. this means the moat for AI 'trillion club' is gone. In fact AI virtually guarantees that there is no execution-moat left anywhere, definitely not in code or that engineer with knowledge about that obscure mechanism. without the moat most of the sw ecossytem's margins would shrink (as they should).
Ironically enough the only moat left would be what you can buy from Washington.
This doesn't account for the fact that review is still a bottleneck, engineers understand much less of the code they're shipping and there'll likely be tech debt they'll have to unwind in the future.
The token-to-feature line is hard to draw partly because most orgs are measuring the wrong end. Tokens burned is an input. The output that matters is something like "PRs merged that wouldn't have been merged this quarter otherwise" minus the regressions those PRs caused 6 weeks later.
If Uber is tracking the second number, I'd love to know what they're seeing. If they're not, the spend will obviously feel unjustifiable, every line item does until you connect it to a unit of work.
I really don’t understand on the customer side of B2B why so many companies actively encouraged AI tooling costs.
I can understand it from the side of the companies selling tokens and AI hardware. I don’t understand the race to spend more on internal tools.
I’ve been sitting around waiting for my company to buy a number of necessary bits of tools. They cheap out on every solution imaginable. Datadog is too expensive, let’s buy a cheap solution that costs us months of setup time. Configuration management is too expensive, let’s use the free version with no audit trail or dashboard.
But everyone…in the entire company…gets multiple AI tool subscriptions.
I don’t remember investors being this stupid at any other point. I don’t recall investors pressuring my company to use blockchain or NFTs.
The logic is quite simple. Management thinks that AI can improve productivity, but knows that there will be some resistance and some learning curve. So they force people to use it so that people can 1. develop their skills and workflow and 2. find out where it is useful 3. find out what needs to be improved to make it useful.
As a more obvious example consider that cars were just invented and the post office management thinks that they could improve performance of letter carriers. But right now cars are slow, break down a lot and there isn't much infrastructure for them. Lots of letter carriers will (rightly) think that it is a waste of time because they need to get in, stop, park between every house and they break down so often it isn't worth it and half of their route is unsuitable for a car anyways. But if cars are forced for a while they will find out what routes work well for cars and which don't, improve the cars and related infrastructure to make cars more effective and other improvements to unlock more productivity.
So yes, right now management is wasting money on cars and gas for no increased productivity. And yes, measuring how much gas each employee uses and encouraging to use more is obviously stupid in isolation. But the idea is to force adoption to iron out the kinks and find out where it can improve productivity. It is basically funding a research project.
IMO, the root of all of this is the almost total inability for most managers, and most eng orgs to measure individual engineer output in any useful way. And in particular in a way that lets you reliably compare engineers to each other.
Despite decades of the industry telling itself that we "pay for performance" or whatever, that has never been the case because we can't really measure performance very well. Where I have seen it done ok (not great, just ok), it was massively labor intensive and did not last, and was only done fully when considering promotion.
So, as you observe, now we have some new technique that managers are sure will increase performance by 50+%, if only people would use it. They can't just raise their expectations of performance by 50%, because they can't measure performance to within 50%! So, they measure the thing they can: token consumption.
That would be sensible if you were a car maker and not in the business of delivering posts. You’re tanking your core business, just to do what the developer of the tool should be doing (ensuring the usefulness and reliability of the tool).
I’m all for a trial run, but it needs to be done like any research experiment. With a goal and measurements along the way. Not by going blind and hurting your workers/customers.
How will companies like Uber continue to fund the “research” when the budget ends up burning 3x faster than predicted without being able to observe measurable gains?
Nothing that C-execs and management advocates for has made any sense for a long time now. If this is the first you're starting to question it all, I must ask what rock you're sleeping under because I desperately need a really good nap...
Strategy - just doing what your friends on the golf course are doing.
The number of times I have been told "oh I talked to so and so and they are having SUCH a good time using X" and then three years later "oh I talked to so and so and they got rid of X as soon as they could, we should switch!"
Or the management all read the same article in PC Magazine and lo! the next day did orders come down to implement said article, regardless. Waterworld, rowing the Valdez. Some years of this usually results in some number of half-baked or half-implemented systems scattered about production, and who knows which if any are actually used, or how much stink there will be to shutdown something unpatchable. Like why are there two wiki engines, sharepoint, three different database servers, …
I surely remember everyone does SOA, everyone does NoSQL, everyone does Hadoop, everyone does microservices, everyone does kubernetes,....
Not with the same pressure as everyone in the company (literally everyone, regardless of the job role) has to burn AI tokens, and attend forced AI workshops, still it is always running after the next new shinny.
Nobody wanted to admit that they had no idea how "AI" was going to help but nobody wanted to get left off the hype train...so they tasked their engineers to figure something out...by just asking them to spend as much as possible (As I explain this it just sounds stupider and stupider). Of course, spending willy-nilly is not a good way to find a profitable (or smart) idea, but that's a problem for future company bottom line.
I like the qualification of "large-ish". When someone says code is not the bottleneck, I assume they work for a large company. To be fair, at a start-up developers think they code at least 4 hours a day, but it never averages that either.
I wish they would spend some of that on the help/support function within their apps. Whenever I have a problem on Uber, it feels like a never-ending maze to figure out how to get any support, and I consider myself versed in navigating unseemly UI, I can only imagine how much it might frustrate people who struggle navigating apps.
Any idea why their help function seems to impenetrable and if AI might help with it?
It might be. It also lets my loyalty go slowly by slowly. Monopolies work until they collapse.
If you can't tell, it frustrates me so much. I wonder how the internal culture of Uber changed when it went from almost zero interest rates to now trying to make lots of profit.
My friend said he realized Uber can just rely on a steady stream of people either growing up or getting laid off and trying to make a quick buck, so they can treat their drivers poorly as well.
I'm not sure what's happening, just know support may be a lot simpler and cheaper to address than nothing, or at least in the medium to long term, but maybe not?
It feels like maybe the wheels are starting to fall off the AI hype train. I expect complete collapse once people start figuring out that the numbers on all this don’t make sense. I’m looking for investment portfolios that will weather that storm. If you are reading this and have a similar curiosity, this is a great place to start.
I've been thinking that for years about various sources and the bubble stubbornly refuses to pop on a convenient timeline so I'm falling back on the adage "time in the market beats trying to time the market". Index funds and chill is much more relaxed than trying to determine who's actually going to survive the AI bubble popping.
this is the reason I refuse to budge from my index portfolio besides small 'play money' ventures. My investing philosophy is basically by the time it hits wire you dont know what portion of it factored in. especially in age of AI and automation IMO alpha will vanish faster as anyone can code up eqv of a bloomberg terminal themselves. So all thats left is how do you manage downturns, when market heads down 30% but your handpicked stocks go down 60% you need to have enough wherewithal to hold through bad times. this is where true test of faith comes, I believe I would be cowardly and sell out at wrong time. So its best to just hold index and market sort the bloodbath out itself.
Outside of tech, most companies have already begun pulling back and cancelling their AI spend. (The trend actually started last year.)
Most of the time, the tasks that AI would take over aren't the bottleneck in the business process, so having AI do something faster isn't very useful. It's definitely not useful enough to justify spending more than two digits a month on a recurring subscription, but the price point at which AI is a viable product is far below the price point necessary to sustain even a single AI company.
I am not sure how uber is operating internally around the use of tokens but if they actually shipped features faster than before then it is still a win. if they learn that users don't want these features or want a different version of it; they have learned this new knowledge faster than they would have if they manually coded those features, which means in principle you should be able to iterate faster. but this will collide with creative ceiling that humans exhibit in a span of time and on top of that uber is prioritizing spending money on tokens over humans which seems like a mistake. you need humans for creativity.
I used Uber for the first time in like 8 years recently, and near as I can tell it's the exact same thing it was. What features are they even adding at all much less that anyone cares about? You ask for a ride to a place, a driver shows up, money is exchanged - the end?
Sometimes things are actually just finished. They don't need to treadmill.
They have to make it easier for ubereats orders to have parts of their bill attached to a corporate card and half attached to someone's personal card and be able to split the invoices!
This is "Feature Factory" thinking, and it is usually not ideal. Feature count is a poor metric to optimize against. Instead, ROI and delivered customer value should be the focus of product development investments.
This argument is funny because you could have said the same thing 4 years ago: Uber still picks you up just as it did years before that, so what did all those millions spent on developer salaries get them?
Uber’s business is relentlessly confusing for people who think it’s a simple app to send an alert to a nearby driver to pick you up.
Uber operates at a scale where there are no trivial problems because even small changes can impact hundred of thousands of customers. They can also justify spending time and money on new features that only 0.1% of customers might use because 0.1% of their customers is a very large number.
Uber also has to maintain thousands of region specific rules and features to be able to operate globally, and they do it all in the same app instead of having specific regional versions (which would be a terrible user experience for frequent travelers). That alone is a ton of work the end user will never see but is core to their operation.
This is not just me saying it. The Uber president himself says it in the article.
> so what did all those millions spent on developer salaries get them?
There was no doubt about what these developer salaries got them. It was to keep Uber stable and running in thousands of jurisdictions with varying rules/regulations.
The idea of using AI was (I hope) not just to replace developers for this purpose but to also ship features/products beyond what was already being offered. It has however not panned out as these CEOs/execs thought it would.
> They can also justify spending time and money on new features that only 0.1% of customers might use because 0.1% of their customers is a very large number.
And what are those features exactly? Because even the President of Uber doesn't seem to know:
"“That link is not there yet, right? I think maybe implicitly there is more that is getting shipped, but it’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25 percent more useful consumer features,’” said Macdonald."
The budget allocated to AI for the year has been wiped out in 4 months.
> Eats for the way - your driver picks up a takeaway for you to eat while they drive you to your destination.
This seems like the kind of terrible idea that an LLM might have come up with. I'm pretty sure most drivers do not want people eating (especially a whole meal) in their car, and I can't imagine a lot of instances where you're calling an Uber and don't have time to get yourself food, but don't mind waiting an extra 10 minutes for the driver to detour, find parking, and wait for your food.
Not to mention what anyone who's worked in an office with a shared kitchen can tell you - the smell getting into a car where an indeterminate amount of people have eaten different meals. Like climbing into a food court dumpster.
> I can't imagine a lot of instances where you're calling an Uber and don't have time to get yourself food
Recently I got a car to take me to the train station and picked up food on the way. Seems pretty common to me. Of course, I didn't need or want it charged as a premium feature in the app.
This seems like the doom of all tech companies that hit a single kernel of a good idea, hire a big development team to build it, and then, once it's running well and making money, leadership looks around and sees this big body of developers, product managers, project managers, QA, and management tree, looking around for something else to do. Then, instead of saying, "Let's find the next big thing to do," they say, "Cram dozens more things into the thing that already works. Anything you can think of, spin up a team of 10 to bolt it onto the main product. Move things around to make everything fit. Run experiments on users to see if this new crap moves the metrics. A/B test to see what we should keep and what we should silently remove next update. Attach this other company's product that we just bought."
In a few years, what do you end up with? The modern version of every single fucking app we use today.
Well, travel booking is one of those things every company wants to get involved with because it's just straight referral fees. I get advertisements to book travel through my phone company (T-Mobile US) and a slew of financial services companies.
If it's easy enough to add to the app and sticks around for a while, it may well be profitable even if only a small percentage of customers use it or even realize it's available.
There's probably tons of backend projects going on, expanding in countries, payments, complying with regulations, effeciency and reliability projects. They also do food delivery. There's a whole engineering team to support
I wonder how hard would be spend maybe 50% of that budget and build a farm of GPUs to run the models locally. I guess it would be a lot of work, but it would be interesting to see how it compares in terms of performance and cost.
They'll expect fewer people to do more with even less, while "more" is continuously increasing.
When I say "more", I mean that the deluge that engineering teams deal with comes from two sources:
1) the business side of companies - marketing, sales, solutions teams, etc.
2) outside actors, mainly security threats
The first source can now move to generate work for engineering faster than ever. They expect the nerds to do what they're told and get the features out now. The more features, the better the product, right? The saving grace here is that they're bound by the same management concerns that engineering has. There's only so much money that they themselves can throw at generating more work for engineering teams, and that might also come under scrutiny from management, so that acts as a brake.
The second source has no such brake, especially not with security threats. Either there's good money to be made by holding company data hostage, or there's an endless supply of resources (read: nation-state resources) dedicated to the effort to attack the company's digital assets. And of course, they're using AI to enable this, just without the "but what about the shareholders!?" handwringing.
If you aren't very, very careful with your token cutting, you're going to put yourself at a disadvantage against that second group.
Didn't google say that AI had increased their company's productivity by 10%? if that's the case, then how can they justify spending 50% to 100% of wages on it?
I do find it to be true that with coding agents the famous quote from Jurassic Park goes through my head multiple time a day
"our scientists were so preoccupied with whether they could, they never stopped to ask if they should.
I've now come to the realization that if I'm having an llm work constantly all day writing code for me i'm probably doing something wrong as I'm no longer focusing on the core issue itself.
I may be in a minority here in that I write code to augment my self and not to ship to others so I can tell very quickly if I'm just gold platting something or if i'm actually delivering real value to my trading or risk management.
I never took tokenmaxxing to be about improving productivity directly; mundane feature work that comes out of it is just a side effect. I always saw it as a race between these big tech companies to get a generational advantage by being the one to discover the way of the future, with respect to harnessing AI to actually and truly automate software development.
EDIT: whoa, I used "way of the future" as a reference to Howard Hughes in "The Aviator", not this Way of the Future religious organization thing I just stumbled on; no intended reference there.
My impression of companies pushing AI so heavily is that they are basically being forced to do it by it merely existing. Imagine if AI really is as powerful as it is suggested to be and you didn't jump on the bandwagon. Then you would be behind. So by it existing and other companies using it, you have to as well because even if it turns out to be a failure at least everyone else will have failed too and you are on an even playing field.
I feel like this is kind of like the gambler's risk of "going bust" but applied backwards?
Feels like it would be better to spend just enough so that you have the capacity to scale up IFF LLM's end up being a big deal. You spent less than your rivals (who were competing for a supremacy that never came), but you have saved more "dry powder" to compete against them for the more likely future. The only future you exclude with this strategy is the "LLM Supremacy" future, which you only had a 1/(number_of_players) chance of winning anyway. :]
I think the real reason for the spending is that the scent of LLMs in the air causes stock values to go up. And even if everyone knows it does not make sense, they still want "NUMBER GO UP", and so they will spend more money to excite the amateur and professional investor class.
You could kind of use that line of thinking to justify spending on anything.
There is a fad every few years that uses similar thinking to justify spending. Sometimes they work out sometimes they don't. But they never work out as good as the most optimistic predictions.
I commented on wasteful AI spending, and my father immediately started talking about how he "went through that a couple of times" as a manager at an oil company.
The unusual thing is perhaps how global and cross industry it seems.
> Sometimes they work out sometimes they don't
Genuinely asking: for which fads was it actually beneficial to jump in during the hype phase? Was there ever anything so critical that there was some huge disadvantage if you didn't adopt it right away?
ETA: I suppose the complicating factor, at least for B2B, is "customers demanding $fad", particularly when the purchasing decision makers don't actually understand what $fad is (e.g., "cloud", "blockchain", "ai", ...). If you don't become "$fad native" right away, you lose the Dunning-Kruger segment of the market.
That's what throws me off about the AI hype. I've lived through plenty of other hype cycles, but none had such rapid adoption as this during the hype phase. Usually its a bunch of early adopters and a few doomed startups, with the big established (non tech) companies never fully jumping in, and then by the time they might have considered it, the fad died.
Even "cloud" which did stick around and actually did pan out, didn't see such immediate adoption during the hype. There were a lot of companies that stayed on-prem for a long time, many which still are, and none of them imploded for not jumping on the hype.
Why is the FOMO so strong with AI this time around? I don't ever recall being told "spend as much money on AWS as you possibly can!" during the cloud hype...
because the tech industry is at a hole is out of creative ideas. That why ever seen so many hype cycles since the end of zirp - and why the only innovation we're seeing is on ways to squeeze more money out of people. and AI was sold as a salary cutting magical money machine... AKA one more hype to jump on.
except this one a isn't making anyone rich besides Nvidia
> because the tech industry is at a whole is out of creative ideas
This isn't true. However the tech industry is out ideas that apply to many many people and scale well.
Most people need a word processor at some point in their life (if your school doesn't make you write at least one paper on a word processor then your education failed - you might never do it again in your life but it is still an important skill), but those were already powerful enough in the 1980s, and the 1990s solved most of the usability issues.
Robotic vacuums can get some more innovation, but the obvious next steps are unsolved problems (I want it to pick up before it cleans) that may not be solvable for a reasonable price.
There are however a ton of niches that could use more technology. However because they are niches they don't scale. You can make millions (gross profit) if you can solve their problems, but not tens of millions - this is enough to get your personally a nice lifestyle if you run or work for such a company (think a 3-5 person company), but even if you could interest an investor there isn't enough for them to skim off any profit and still make money.
Not all niches that need tech are that small. There are a few large ones, but they are hard to find (if they were easy someone would have done it already). There are also a lot of what looks like large ones that either are not large, or are not large enough to pay for the investment needed. There are also some medium sized places that tech can help. Once in a while there are even tiny places where someone can make a difference (but generally this means you do the thing as your business and tech as a hobby after work)
> Robotic vacuums can get some more innovation, but the obvious next steps are unsolved problems (I want it to pick up before it cleans) that may not be solvable for a reasonable price.
Roborock has already released a vacuum that does this. From what I've seen it's limited, but it seems to work for the things it can pick up.
Sears, the obvious choice, could have been Amazon. There are plenty of other less extreme examples of defunct brick-and-mortar stores who were killed by internet competitors (often, Amazon).
Usually, someone extracts genuine value at the expense of everything else IMO. And the enshittification continues apace.
Only lemming spending. Where the first few lemmings have to justify it some other way.
Not on anything but on some things like “Nobody Ever Got Fired for Buying IBM”
Yeah this is right. Its about taking the throttle off of experimentation hoping some team suddenly starts shipping a years worth of features in a week and responding to strategic customer demands in near real time. Then copying what works out across the org. (and probably downsizing)
What those organizations miss is that they drown their teams in organizational red tape. The way to the future isn't tokenmaxxing - it's cutting back on process noise.
It was very easy to see before this started that the models were just repetitive inflators.
It’s not that they should “scale back” their use as much as the metric should be improvement/tokens. Tokens used is a denominator in any worth calculation.
> by being the one to discover the way of the future
This is my understanding too. The underlying assumption is that action leads to information, iterations lead to enlightenment. So from an org's point of view, tokenmaxxing means encouraging everyone to explore as much as they can. Of course, token volume should not be the only metric - tokenmaxxing is just a catchy phrase.
> action leads to information, iterations lead to enlightenment
So doing something (action) creates something new (more information), and iterating on that new information leads to the realization there is nothing new left to be learned with that information (enlightenment). Is how I'm interpreting that.
To the extent that figuring out the true automation will lead to layoffs, that knowledge will quickly spread as the laid-off knowledge workers are available for hiring.
Like many corporate behaviors I believe we tend to rationalize them too much. Most job posts for management ask for previous experience with "implementing AI". Managers don't want to be left out any more than devs. Devs rewrite everything in the language-du-jour, managers max out tokens.
My hot take: The way of the future is local LLMs.
The AI equivalent of the PC revolution isn’t quite here yet, but it’s the only way forward.
I agree. We don't need a nondeterministic 10quadrillion vector model. We need an deterministic expert on our narrow business. Something small, that can be run on the 2026 version of the spare PC under the CTOs desk.
It will always be worse compared to a centralized approach where hardware utilization can remain high. Except in case which demand low latency which most development things do not need. It's okay if it takes an extra 100ms for a code review to take place.
It's little different from the mainframe/mini computer to the PC: Huge servers and resources will be better--no dispute there--but good-enough will be achieved using local resources.
There are privacy and general de-centralization reasons to prefer this outcome, even though most AI and cloud-first tech companies don't want this.
How long it will take us to get this point is a different matter.
By that logic shouldn’t streaming video games also be centralized?
Yet startups keep trying it and failing. Turns out users actually want exclusive access to that hardware to have a smooth experience. The tradeoff has always been between faster exclusive hardware or slower but cheaper shared hardware.
If local hardware can’t beat shared hardware on performance then something’s wrong? Either it’s because the providers are charging wildly below cost or because local hardware just hasn’t needed to catch up. Maybe it’s both.
RAMageddon has caused local models to lag far behind hardware for a long while now.
Agree and I have wondered behind close doors if this is not the mental model. You need to spend money to see what is working this was simply a way to see that.
This is insightful. It does sort of feel like the search for the Northwest passage.
Eh, I also saw it as a rather blatant attempt to undermine the bargaining power of the Software Engineer (which had grown to insane proportions over the years) - both in working conditions and raw cash money.
In many cases it really didn’t/doesn’t matter if the AI automation actually works, just that people think it could - and hence leave money on the table.
> insane
Not sure if you mean this in a good or bad way.
Good/bad entirely depends on if you are management/shareholders, or the engineer in question. It was pretty nuts either way though eh? End of an era
[flagged]
Now feels like a very good time to be a small team of experienced developers who can largely work on stuff by themselves and not a corporation of hundreds of developers of varying abilities all now trying to show how much code they can generate and how many tokens they can burn.
100% - I think that is also part of the divide you see online. Devs who work on massive codebases w/ 100s of engineers and see the bugs the LLMs create vs devs who work on smaller codebase w/ small <5 person team.
It's a tradeoff.
Generating a feature that is 90% correct in a tenth of the time is a reasonable tradeoff if you're trying to gain traction.
Generating a feature that is 90% correct in a tenth of the time, risking a multi-billion-dollar business, is a terrible tradeoff.
I think it's rather:
Small teams building continuously get to write features that are 90% correct in a tenth of the time.
Big enterprises get to write features that are 90% correct barely twice as fast, because all of the bottleneck lies elsewhere. They also spend more on AI per user because of the internal dynamics pushing people to adopt AI irresponsibly. They can correct the 10% of errors slower than small teams because of bureaucracy, increasing the cost of errors that show up in the product. Furthermore, they have less to gain from a given amount of speedup because they had plenty of engineering velocity anyway compared to small teams.
I don't think big enterprises will start winning from AI technology until AI truly can automate almost everything in a company and let said company outproduce competitors by burning tokens alone. That's nowhere near possible right now.
I don't think "90% correct in a tenth the time" is really the tradeoff. For a well-specified task, it's closer to 100% correct in 1/100th the time. But that undersells the time required to generate a good spec.
For under-specified tasks, it's not really accurate to talk about "correctness," because the machine isn't psychic. I would suggest that given a high-level feature request like "add streaming support" it's more about acceptance probability. In a well-structured and well-documented codebase, and a reasonably sized feature request, there might be an 80% chance it will generate something which is 100% acceptable. But there's about a 99% chance it will generate something which is acceptable after 1-2 revisions.
[dead]
Even small teams are not immune from the AI push. I work in a small <10 people with 3 devs. I'm the only one not using AI while receiving push that I'm a problem for not. To the point, VP of company says if we don't all start using AI we'll be out of business because other people will. :face-palm:
You don't use AI at all? What is your main justification for not?
I don't like outsourcing learning and prefer to do the actual learning. Just like taking notes via recorder vs writing them down during a lecture causes you to not remember the information as well. Allowing something else to create your code tends to mean you're not going to remember/know the how/why it works and you just accept it works and moves on. Eventually you don't know how any of it works and you're totally dead in the water without the LLM. My understanding of the code I write is much more valuable than some arbitrary deadline that means nothing
Do you also inspect and study what assembly code your program was compiled to?
// Obviously LLMs are non-determenistic etc and it depends on your domain, but your VP's point 100% makes sense if you folks are trying to cook up another demo-CRUD apps to convince investors for another funding round
I keep hearing this argument on HN. Yes, if performance is something you care about you totally look at the disassembly. You don’t even need to write assembly, just be able to read it. By now I assume that programmers who argue they don’t need to understand and review what the LLM generates for them are simply not that skilled and don’t realize that skilled programmers do this.
If you're just building a non-security-sensitive frontend to validate market traction, it makes total sense to go full AI. The commenter working at a 10-person company getting flamed for not using AI to iterate aggresively against competitors has a super valid point.
Validation methods will evolve to accommodate human laziness. Insisting on doing it the hard way is no different than the old-timers who used to claim engineers 'weren't skilled' if they didn't know how to use punch cards.
Ad hominem. Skilled programmers != old-timers calling others unskilled. Who does fuzz testing, automated integration tests, etc already, that LLMs need as a test harness? It’s not the vibe coders saying “nobody reads assembly anyway” as an argument.
I simplified this argument to highlight that we’ve seen this exact pattern before.
I'd definitely say IntelliJ improved my productivity. No one cared. If anything, management viewed it as a nuisance as they had to deal with the license.
Now there is demands to justify not using AI like this, but people don't care about details. Which AI tool I use apparently doesn't matter at all, even if there are presumably productivity differences between them.
Edit: typo
Seems like you should have a justification for introducing a new tool instead, no? There are thousands of other tools they're also not using, but we're not asking about those.
I'm not the same person, but I don't really use it. I write better code than an LLM, and I do it faster than the time it takes to review and correct the LLM's bad code. The tool has no value to offer me.
!RemindMe 2 years
To be clear I meant it's good to be small team on the assumption that we're all using AI... I honestly can't imagine not using AI in May 2026.
He could be wrong, but in my opinion, he is probably right. Others are, and will be using AI.
I think the jury’s still out on how big an impact AI will have on overall, average productivity. But it’s definitely a productivity boost for someone who’s capable of writing code without it. If you want to be super conservative, don’t even have it write any code. Use it to search through existing codebases, review your code, find the root cause of bug reports, evaluate pros and cons of alternative approaches, etc. etc. You’re really missing out by not using it at all.
Here’s a concrete example of conservative AI usage: I use Claude to vibe code my nvim config. Now, who cares if my nvim config is AI slop? What’s the worst that can happen? Nvim works for me now way better than it ever did when I was limited by the time I was willing to spend configuring it manually.
What has been the end result of all the tokens companies are burning?
Where does it show up in quarterly results?
I can’t see how it’s sustainable just based on “this feels more productive”
A friend of mine added some pretty extensive iOS UI tests to a keystone feature hit by millions every month. They'd been kicking the can down the road for years, trying to fit it in their roadmap, and with Claude running overnight they were able to bang out the whole suite in a week.
I'm not sure how it would show up in quarterly results.
I see these kinds of stories here a lot, and I'm curious whether they reflect a steady stream of need for AI coding, or whether a lot of companies have a burst of AI-appropriate coding work now that the technology is available and then will have a smaller need going forward.
Is it like the stereotypical dad who rents a power washer, powerwashes every exposed surface on his property, and then doesn't need to do any powerwashing for a few years; his neighbor who gets an Instant Pot and uses it for every meal for a month, then sees it gathering dust when the family gets tired of pressure-cooked stews; or like their neighbor who gets a microwave oven and uses it multiple times a day for decades?
I guess only time will tell.
So far where I work its the Instant Pot, at least for the non-devs. We rolled out Claude & Cowork to the masses after a brief pilot. It was about a solid month and a half of heavy usage and then suddenly usage fell off a cliff. Once it stopped being a cool new toy, people just didn't find a use for it.
A few mundane things got automated, but these were just back office admin type work. Nothing that's going to show on the P&L. Yeah those people now have a little more time for other things, but those other things are also not revenue generating. No FTE got replaced by it so in the end they just paid for a bunch of administrative positions to be a little less busy. Great for the workers who are now less stressed, but almost no impact on the business financials except there's now yet another subscription.
> So far where I work its the Instant Pot, at least for the non-devs. We rolled out Claude & Cowork to the masses after a brief pilot. It was about a solid month and a half of heavy usage and then suddenly usage fell off a cliff. Once it stopped being a cool new toy, people just didn't find a use for it.
Your employer is doing it wrong. You need usage surveillance with sanctions for low/declining use, then people won't stop using it.
Please tell me you're being sarcastic.
It's industry best practice. All the market-leading companies are doing it. Do you think you know better than them?
If there's anything I've learned as a software engineer, it's that agreeing with and defending the ideas of business leaders and Silicon Valley VC influencers proves I'm very intelligent.
this sarcasm is very disrespectful, you're mocking a sizeable proportion of the commenters on this site.
when I quote this comment later, with appropriate attribution, please know that I will be shaking my head and frowning while doing so
> you're mocking a sizeable proportion of the commenters on this site.
some would call this god's work
He is, but he's also describing reality.
That's depressing
He’s referencing practices at Meta and probably others
That’s been my theory - there’s some low hanging fruit in every environment where AI knocks it out of the park. Then complex brownfield reality (coupled with non-technical factors) rears its head and the stunning productivity gains are nowhere near to be seen.
That’s the explanation how you can have both the anecdotes of amazing AI productivity and rigorous studies showing anything from actual loss of productivity to single-digit gains.
I also think in addition to that the increased speed compounds the problems much quicker. And I don't mean bugs. I mean that duplicated code here, that additional state variable to keep everything going there. Not removing things that should be removed because we can work around that, etc.
It's like building a super tall Jenga tower very quickly but laying the bricks much worse than a careful player.
I think this is directionally right.
The code AI produces is not created equally, not even close.
I've found it to be that people simply go back to their old ways. Many people are busy and unless they're being actively pushed to change it they feel they only have time to do things the way they have been doing them.
Using AI requires skills to know how to use it, particularly agents then it requires the time to build an improved way of doing things.
I think of it like giving someone excel in the beginning and expecting they know how to use it, when the rest of their team doesn't, they don't have the skills to know how to use it and how it can benefit them.
> or whether a lot of companies have a burst of AI-appropriate coding work now that the technology is available and then will have a smaller need going forward
For the product my friend works on, it's definitely the latter. I definitely don't expect this party to last forever.
Nah in competitive industries you need to build features and out compete people and getting AI to do that whilst architecting things well due to experience and having time to think more about the important stuff but have a lot of the more boilerplate and simple things ABs plumbing etc handled by agents is great.
When you try to replace your entire brain with AI things are going to go wrong.
Some measures should have real, tangible, concrete numbers; others should have “my friends are saying”/“you are blind if you are not seeing it” vibing.
Everything can be expressed in financial terms- it is one of the guiding principles of the universe. Anyone who thinks they are above or beyond it will rue the day.
Yes but that representation can be over-applied and misleading. For example you can roughly estimate the price of the entire earth. What does that number mean though? Well, nothing, really.
I hope that you never loose a child and remain so naive.
In my limited experience with using agents to create tests it tends to code the tests to the existing code instead of ensuring the correctness from a spec. Great for regression testing but still limited in effectiveness for catching existing issues.
Over here our CTO replaced Intercom with an internal equivalent that costs less than $20 / month to run, haiku and sonnet support agent costs included. In less than a few weeks, in his spare time.
It wouldn't, at least not directly. That's why it wasn't done pre-AI.
Even if Uber really did double developer productivity, would it translate to quarterly results?
Ultimately they make money selling rides, not selling software. The Uber app is mature and adding new features is unlikely to significantly increase sales.
Writing 2x more code doesn't translate to 2x more revenue unless it results in 2x more rides.
> Even if Uber really did double developer productivity, would it translate to quarterly results?
It would if it meant they then fired half their software engineers, which is the ultimate goal.
Even if they lower costs to 0, it means nothing
Topline growth matters more than costs.
Mathematically - There is no limit to topline growth But cost cutting has a ceiling, which is the current costs.
You can only make(save) so much money with cost cutting
Topline growth is upper bound by total energy in the universe though.
Topline growth is more realistically bounded by the number of people willing to take an Uber at a given price. You can't expand forever.
Lowering costs to run the infra would show up as increased profits without any change in rides.
Do you really need AI for that? Seems like the thing any existing engineering team could do if it was prioritized.
How much is infra spending compared to revenue/other expenses? I honestly can’t imagine it’s that high, they’re not running something like Netflix or YouTube… But maybe I’m underestimating it.
Would it? How much of Uber’s cost is software infrastructure vs paying humans?
In the big red number shown after revenue where profits used to be.
Probably shows up in OpenAI and Anthropic quarterly reports. I have to wonder if that was the point.
Advancement in AI research seems to be the only thing at this point.
> Where does it show up in quarterly results?
Standard answer is "companies that are not seeing significant gains from AI just aren't AI-ing hard enough, trust me bro".
Anthropic's annualized run rate is >$40b according to outside reporting. AWS hit that by Q4 2019. There were still debates on public cloud vs on prem at that time, but by late 2019 public cloud had facilitated the creation or adoption of entire categories of software within SaaS and PaaS, not to mention consumer internet businesses like Uber and Airbnb. The net impact of AI coding tools is far more ambiguous in comparison.
The profitability comparison is fraught but worth noting that by then AWS was already extremely profitable.
Amazon was AWS first customer. I sometimes feel if the AI companies promises of their models replacing all SWE, they would be a top 3 product/service in any number of businesses and not selling shovels to miners.
But they do use their models internally quite a bit. Claude code head talks about how he doesn't write any code himself anymore and just uses agents to do everything. So they are using it themselves.
> Amazon was AWS first customer
It wasn't. The retail business took years to move to AWS. They could not even be described as early adopters of AWS.
”AI models will replace software developed in 6 months”
— CEO of Anthropic, the employer of over 2000 developers, over 6 months ago
This is an apples to some non-food item comparison.
AWS has so many analogs. It’s not as novel. Renting vs buying a home/car/anything is essentially what AWS brought.
Anthropic will not have as many husbands as they think by next year: https://xkcd.com/605/
I actually think Chinese models have already popped the bubble, we just dont see it yet. the only way to justify AI IPO market cap is basically if they get to hold most sw industry code hostage and then token-flate budgets to collect AI tax. short of that AI expense would very quickly mean revert to model + some margin. this means the moat for AI 'trillion club' is gone. In fact AI virtually guarantees that there is no execution-moat left anywhere, definitely not in code or that engineer with knowledge about that obscure mechanism. without the moat most of the sw ecossytem's margins would shrink (as they should).
Ironically enough the only moat left would be what you can buy from Washington.
Agreed.
We have to wait a whole year (sigh) since firms generally wait a whole financial year to do critical reviews
This doesn't account for the fact that review is still a bottleneck, engineers understand much less of the code they're shipping and there'll likely be tech debt they'll have to unwind in the future.
The token-to-feature line is hard to draw partly because most orgs are measuring the wrong end. Tokens burned is an input. The output that matters is something like "PRs merged that wouldn't have been merged this quarter otherwise" minus the regressions those PRs caused 6 weeks later.
If Uber is tracking the second number, I'd love to know what they're seeing. If they're not, the spend will obviously feel unjustifiable, every line item does until you connect it to a unit of work.
Goodhart's law strikes again. Stop giving your engineers token-burning quotas or they'll burn tokens.
I really don’t understand on the customer side of B2B why so many companies actively encouraged AI tooling costs.
I can understand it from the side of the companies selling tokens and AI hardware. I don’t understand the race to spend more on internal tools.
I’ve been sitting around waiting for my company to buy a number of necessary bits of tools. They cheap out on every solution imaginable. Datadog is too expensive, let’s buy a cheap solution that costs us months of setup time. Configuration management is too expensive, let’s use the free version with no audit trail or dashboard.
But everyone…in the entire company…gets multiple AI tool subscriptions.
I don’t remember investors being this stupid at any other point. I don’t recall investors pressuring my company to use blockchain or NFTs.
The logic is quite simple. Management thinks that AI can improve productivity, but knows that there will be some resistance and some learning curve. So they force people to use it so that people can 1. develop their skills and workflow and 2. find out where it is useful 3. find out what needs to be improved to make it useful.
As a more obvious example consider that cars were just invented and the post office management thinks that they could improve performance of letter carriers. But right now cars are slow, break down a lot and there isn't much infrastructure for them. Lots of letter carriers will (rightly) think that it is a waste of time because they need to get in, stop, park between every house and they break down so often it isn't worth it and half of their route is unsuitable for a car anyways. But if cars are forced for a while they will find out what routes work well for cars and which don't, improve the cars and related infrastructure to make cars more effective and other improvements to unlock more productivity.
So yes, right now management is wasting money on cars and gas for no increased productivity. And yes, measuring how much gas each employee uses and encouraging to use more is obviously stupid in isolation. But the idea is to force adoption to iron out the kinks and find out where it can improve productivity. It is basically funding a research project.
IMO, the root of all of this is the almost total inability for most managers, and most eng orgs to measure individual engineer output in any useful way. And in particular in a way that lets you reliably compare engineers to each other.
Despite decades of the industry telling itself that we "pay for performance" or whatever, that has never been the case because we can't really measure performance very well. Where I have seen it done ok (not great, just ok), it was massively labor intensive and did not last, and was only done fully when considering promotion.
So, as you observe, now we have some new technique that managers are sure will increase performance by 50+%, if only people would use it. They can't just raise their expectations of performance by 50%, because they can't measure performance to within 50%! So, they measure the thing they can: token consumption.
That would be sensible if you were a car maker and not in the business of delivering posts. You’re tanking your core business, just to do what the developer of the tool should be doing (ensuring the usefulness and reliability of the tool).
I’m all for a trial run, but it needs to be done like any research experiment. With a goal and measurements along the way. Not by going blind and hurting your workers/customers.
How will companies like Uber continue to fund the “research” when the budget ends up burning 3x faster than predicted without being able to observe measurable gains?
Nothing that C-execs and management advocates for has made any sense for a long time now. If this is the first you're starting to question it all, I must ask what rock you're sleeping under because I desperately need a really good nap...
I knew they were stupid, I just didn’t know it got way down to this level
Strategy - just doing what your friends on the golf course are doing.
The number of times I have been told "oh I talked to so and so and they are having SUCH a good time using X" and then three years later "oh I talked to so and so and they got rid of X as soon as they could, we should switch!"
Or the management all read the same article in PC Magazine and lo! the next day did orders come down to implement said article, regardless. Waterworld, rowing the Valdez. Some years of this usually results in some number of half-baked or half-implemented systems scattered about production, and who knows which if any are actually used, or how much stink there will be to shutdown something unpatchable. Like why are there two wiki engines, sharepoint, three different database servers, …
I surely remember everyone does SOA, everyone does NoSQL, everyone does Hadoop, everyone does microservices, everyone does kubernetes,....
Not with the same pressure as everyone in the company (literally everyone, regardless of the job role) has to burn AI tokens, and attend forced AI workshops, still it is always running after the next new shinny.
I think most of the previous trends were less widespread and some of them (like kubernetes[1]) are sensible default options.
We are seeing shoe companies pivot to AI. They didn’t do that with Hadoop or NoSQL.
[1] Even some homelab folks sometimes go straight to kubernetes even though it’s technically overkill.
Nobody wanted to admit that they had no idea how "AI" was going to help but nobody wanted to get left off the hype train...so they tasked their engineers to figure something out...by just asking them to spend as much as possible (As I explain this it just sounds stupider and stupider). Of course, spending willy-nilly is not a good way to find a profitable (or smart) idea, but that's a problem for future company bottom line.
There's also the issue that in any large-ish org, code production is hardly ever the bottleneck.
I like the qualification of "large-ish". When someone says code is not the bottleneck, I assume they work for a large company. To be fair, at a start-up developers think they code at least 4 hours a day, but it never averages that either.
I wish they would spend some of that on the help/support function within their apps. Whenever I have a problem on Uber, it feels like a never-ending maze to figure out how to get any support, and I consider myself versed in navigating unseemly UI, I can only imagine how much it might frustrate people who struggle navigating apps.
Any idea why their help function seems to impenetrable and if AI might help with it?
I think this is clearly be design. They don't want to provide support, they want you to give up and let your issue go.
It might be. It also lets my loyalty go slowly by slowly. Monopolies work until they collapse.
If you can't tell, it frustrates me so much. I wonder how the internal culture of Uber changed when it went from almost zero interest rates to now trying to make lots of profit.
My friend said he realized Uber can just rely on a steady stream of people either growing up or getting laid off and trying to make a quick buck, so they can treat their drivers poorly as well.
I'm not sure what's happening, just know support may be a lot simpler and cheaper to address than nothing, or at least in the medium to long term, but maybe not?
It feels like maybe the wheels are starting to fall off the AI hype train. I expect complete collapse once people start figuring out that the numbers on all this don’t make sense. I’m looking for investment portfolios that will weather that storm. If you are reading this and have a similar curiosity, this is a great place to start.
https://portfoliocharts.com/2021/12/16/three-secret-ingredie...
I've been thinking that for years about various sources and the bubble stubbornly refuses to pop on a convenient timeline so I'm falling back on the adage "time in the market beats trying to time the market". Index funds and chill is much more relaxed than trying to determine who's actually going to survive the AI bubble popping.
this is the reason I refuse to budge from my index portfolio besides small 'play money' ventures. My investing philosophy is basically by the time it hits wire you dont know what portion of it factored in. especially in age of AI and automation IMO alpha will vanish faster as anyone can code up eqv of a bloomberg terminal themselves. So all thats left is how do you manage downturns, when market heads down 30% but your handpicked stocks go down 60% you need to have enough wherewithal to hold through bad times. this is where true test of faith comes, I believe I would be cowardly and sell out at wrong time. So its best to just hold index and market sort the bloodbath out itself.
Market makers are not going to let anything collapse, there is not going to be a "storm".
The government and everyone with any money/power are fully invested in keeping the market going regardless of any kind of reality.
"Every American child under 18 with a Social Security number can have a federally recognized "Trump Account," a one-time $1,000 IRA seed deposit"
By doing this every citizen will personally have skin in the game and want markets to continue to rise.
So what you are saying is “this time it’s different”. That’s a dangerous thought to have.
https://www.currentmarketvaluation.com/models/s&p500-mean-re...
Outside of tech, most companies have already begun pulling back and cancelling their AI spend. (The trend actually started last year.)
Most of the time, the tasks that AI would take over aren't the bottleneck in the business process, so having AI do something faster isn't very useful. It's definitely not useful enough to justify spending more than two digits a month on a recurring subscription, but the price point at which AI is a viable product is far below the price point necessary to sustain even a single AI company.
I am not sure how uber is operating internally around the use of tokens but if they actually shipped features faster than before then it is still a win. if they learn that users don't want these features or want a different version of it; they have learned this new knowledge faster than they would have if they manually coded those features, which means in principle you should be able to iterate faster. but this will collide with creative ceiling that humans exhibit in a span of time and on top of that uber is prioritizing spending money on tokens over humans which seems like a mistake. you need humans for creativity.
I used Uber for the first time in like 8 years recently, and near as I can tell it's the exact same thing it was. What features are they even adding at all much less that anyone cares about? You ask for a ride to a place, a driver shows up, money is exchanged - the end?
Sometimes things are actually just finished. They don't need to treadmill.
They have to make it easier for ubereats orders to have parts of their bill attached to a corporate card and half attached to someone's personal card and be able to split the invoices!
This is "Feature Factory" thinking, and it is usually not ideal. Feature count is a poor metric to optimize against. Instead, ROI and delivered customer value should be the focus of product development investments.
> if they actually shipped features faster than before then it is still a win
Depends on the cost
Dupe: https://news.ycombinator.com/item?id=48268871
go ahead
I still get picked up by an Uber the same way. As an end user, nothing has changed for me.
So I wonder what the heck were all those billions of AI tokens burnt on that they extinguished it in just 4 months into the year?
This argument is funny because you could have said the same thing 4 years ago: Uber still picks you up just as it did years before that, so what did all those millions spent on developer salaries get them?
Uber’s business is relentlessly confusing for people who think it’s a simple app to send an alert to a nearby driver to pick you up.
Uber operates at a scale where there are no trivial problems because even small changes can impact hundred of thousands of customers. They can also justify spending time and money on new features that only 0.1% of customers might use because 0.1% of their customers is a very large number.
Uber also has to maintain thousands of region specific rules and features to be able to operate globally, and they do it all in the same app instead of having specific regional versions (which would be a terrible user experience for frequent travelers). That alone is a ton of work the end user will never see but is core to their operation.
This is not just me saying it. The Uber president himself says it in the article.
> so what did all those millions spent on developer salaries get them?
There was no doubt about what these developer salaries got them. It was to keep Uber stable and running in thousands of jurisdictions with varying rules/regulations.
The idea of using AI was (I hope) not just to replace developers for this purpose but to also ship features/products beyond what was already being offered. It has however not panned out as these CEOs/execs thought it would.
> They can also justify spending time and money on new features that only 0.1% of customers might use because 0.1% of their customers is a very large number.
And what are those features exactly? Because even the President of Uber doesn't seem to know:
"“That link is not there yet, right? I think maybe implicitly there is more that is getting shipped, but it’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25 percent more useful consumer features,’” said Macdonald."
The budget allocated to AI for the year has been wiped out in 4 months.
Apparently:
* In App Hotel bookings in partnership with Expedia.
* Travel Mode with suggestions on where to eat and visit when travelling.
* Eats for the way - your driver picks up a takeaway for you to eat while they drive you to your destination.
* Voice bookings using AI and speech to text.
How did we ever live without them!
> Eats for the way - your driver picks up a takeaway for you to eat while they drive you to your destination.
This seems like the kind of terrible idea that an LLM might have come up with. I'm pretty sure most drivers do not want people eating (especially a whole meal) in their car, and I can't imagine a lot of instances where you're calling an Uber and don't have time to get yourself food, but don't mind waiting an extra 10 minutes for the driver to detour, find parking, and wait for your food.
Not to mention what anyone who's worked in an office with a shared kitchen can tell you - the smell getting into a car where an indeterminate amount of people have eaten different meals. Like climbing into a food court dumpster.
> I can't imagine a lot of instances where you're calling an Uber and don't have time to get yourself food
Recently I got a car to take me to the train station and picked up food on the way. Seems pretty common to me. Of course, I didn't need or want it charged as a premium feature in the app.
I have never heard of someone doing that tbh, this is the first time.
You can get any Uber driver to do it for you if you offer to buy them food too and do it off the books.
You just do a ride with two stops, one at the restaurant, and ask them to wait. It's not remotely clandestine or complicated.
I didn't say I doubted it was possible lol, just that I had never heard of anyone doing it.
Holy fuck, aside from the voice bookings, that's some useless shit to spend money building as far as both tokens and salaries go.
Are they profitable yet lol
This seems like the doom of all tech companies that hit a single kernel of a good idea, hire a big development team to build it, and then, once it's running well and making money, leadership looks around and sees this big body of developers, product managers, project managers, QA, and management tree, looking around for something else to do. Then, instead of saying, "Let's find the next big thing to do," they say, "Cram dozens more things into the thing that already works. Anything you can think of, spin up a team of 10 to bolt it onto the main product. Move things around to make everything fit. Run experiments on users to see if this new crap moves the metrics. A/B test to see what we should keep and what we should silently remove next update. Attach this other company's product that we just bought."
In a few years, what do you end up with? The modern version of every single fucking app we use today.
Well, travel booking is one of those things every company wants to get involved with because it's just straight referral fees. I get advertisements to book travel through my phone company (T-Mobile US) and a slew of financial services companies.
If it's easy enough to add to the app and sticks around for a while, it may well be profitable even if only a small percentage of customers use it or even realize it's available.
they are very profitable now!
For context, this is an interview where Uber CEO discussed these ideas:
https://www.theverge.com/podcast/922909/dara-khosrowshahi-ub...
Can't say I am convinced.
There's probably tons of backend projects going on, expanding in countries, payments, complying with regulations, effeciency and reliability projects. They also do food delivery. There's a whole engineering team to support
I wonder how hard would be spend maybe 50% of that budget and build a farm of GPUs to run the models locally. I guess it would be a lot of work, but it would be interesting to see how it compares in terms of performance and cost.
Affordable inference will be around longer if more Big tech companies cap their AI sending.
My concern here is that they'll mix two things:
1) workforce reduction
2) AI spend (reduce tokenmaxing)
They'll expect fewer people to do more with even less, while "more" is continuously increasing.
When I say "more", I mean that the deluge that engineering teams deal with comes from two sources:
1) the business side of companies - marketing, sales, solutions teams, etc.
2) outside actors, mainly security threats
The first source can now move to generate work for engineering faster than ever. They expect the nerds to do what they're told and get the features out now. The more features, the better the product, right? The saving grace here is that they're bound by the same management concerns that engineering has. There's only so much money that they themselves can throw at generating more work for engineering teams, and that might also come under scrutiny from management, so that acts as a brake.
The second source has no such brake, especially not with security threats. Either there's good money to be made by holding company data hostage, or there's an endless supply of resources (read: nation-state resources) dedicated to the effort to attack the company's digital assets. And of course, they're using AI to enable this, just without the "but what about the shareholders!?" handwringing.
If you aren't very, very careful with your token cutting, you're going to put yourself at a disadvantage against that second group.
Didn't google say that AI had increased their company's productivity by 10%? if that's the case, then how can they justify spending 50% to 100% of wages on it?
I do find it to be true that with coding agents the famous quote from Jurassic Park goes through my head multiple time a day
"our scientists were so preoccupied with whether they could, they never stopped to ask if they should.
I've now come to the realization that if I'm having an llm work constantly all day writing code for me i'm probably doing something wrong as I'm no longer focusing on the core issue itself.
I may be in a minority here in that I write code to augment my self and not to ship to others so I can tell very quickly if I'm just gold platting something or if i'm actually delivering real value to my trading or risk management.
You don't need justification to spend other people's money!
Nobody's going to jail.
And here we are.
hot take: token spend can be used a honey pot, especially when compared to what you deliver. spend accordingly!
[flagged]