It says that the self-interest of an agent (e.g. the CEO) is not identical to the interests of the principal (e.g. the shareholders).
Why do I mention this? The CEO being rich is a measurement of the financial success of the agent, but not of the principal. You are rather interested in measuring the success of the principal.
About 2 decades ago, there were multiple studies showing that C-level execs who offshored work/production got 18% higher compensation. I'm sure that the same bizjournals are claiming that C-level execs that "embrace" AI are getting similar compensation premiums over anyone who takes a "wait and see" approach. The quantity of AI appearing in bizjournals reminds me of Paul Graham's essay on The Submarine.
All we can really do is point and laugh. Boards don't listen to workers, and I bet most boards will be okay with a little spending oopsie-daisy because it was 'try shit and see if it works out'
We are since at least the 2000s when the world wide web revolution finally made it possible for Google, Apple, Amazon etc... to become the behemoths they are. And maybe even before that.
And the answer is no, because when chief officers succeed it's because of their genius and forward thinking posture, but when they fail it's none of their fault, so they are shielded in an echo chamber that produces such delusions and psychosis.
Way before that, as usual we can attribute quite a lot of stupidity in corporate governance to Jack Welch. Execs really bought into Welch's schtick wholly, they went to MBA schools praising Welch's management style, read his books, or at least got taught by people who had bought wholly into it.
So much time has passed that I believe truly the current crop of execs don't know any better, they think this status quo is the only way to manage companies. They aren't really wrong since the incentives are there, and they continue to reap rewards from doing it.
Agreed. For a long time Jack Welch was lauded as "The Greatest CEO in History" for burning GE to the ground in a way that made a lot of money (while destroying a company that took 100 years to build and employed 400,000 people).
By the time he died (pretty recently actually, 2020!), it was pretty obvious what kind of legacy he was leaving behind. Which is probably why his family was very careful to keep his burial location secret, presumably to keep people from peeing on his grave.
When tokens get correctly priced, all of the insane over-investment in capital will need to draw back: buying data centers, semiconductors, and politicians.
Even then, it won't be right-priced with regard to actual costs. The environmental impact should have been priced in from the beginning. There seems to be a parallel with subsidizing fossil fuels, under pricing them which encourages over dependence, ignoring the real costs society will pay later.
It rather looks like chatgpt/antrophic enterprise tokens and API calls are too expensive. Competition is quite strong on openrouter.
However, the real problem is running wild with token burning. With parallel agents calling subagents you can burn lots of tokens per minute. Especially with thousands of engineers.
I've yet to see any compelling data about inference being particularly expensive. For local LLM models, that are becoming increasingly viable, it's dirt cheap. The same is also true in image gen world where now even a heavily dated GPU can cheaply and quickly produce high quality images.
I also think the image gen world is a useful analog because there are a million sites, presumably still making money, with markups that are multiple orders of magnitude off their costs. They're feeding off user ignorance that was, at least in part, artificially seeded by implying high costs for image gen back in its day. Though it's possible/probable that the initial training runs were expensive, but that's a one-and-done cost.
We use GH Copilot at work and this week sat for a presentation by GH about optimizing token usage and maximizing ROI on tokens used. Anyone else get this presentation? They didn’t have time for questions because they had to run and give it to the next big enterprise on their list…
They basically said that everything is too expensive, you have to watch it like a hawk. It was as if they poured a bucket of cold water on the room. People were wondering how they could do anything faster with all these strategies. And then “sorry no questions. Bye!”
Interesting. We use Kiro here and looking at the public pricing subscriptions and it's benefit to my workflow, it is clearly a significant productivity increase per dollar spent. And we were told we have a signed a deal that is better than that public pricing. They recently just enabled overages on everyone's account so that people aren't throttled and they are shifting people up/down tiers as required behind the scenes to align with their actual usage.
However when the 'cost' to do something is relatively flat the cost/benefit analysis is going to depend on the value of the person being enabled. Someone making $60k a year using AI to gain a 20% output improvement may not be worth the cost but someone making $160k a year would.
The ones I’ve stumbled upon seem to be: switch models based on task complexity, use tooling like ASTs and compression, disable unused MCPs, compact often, be verbose with input to give clear guidance…
I don't think compacting often is good for saving money. It generates more output tokens and then the input is no longer from cache, which is priced differently...typically very differently.
We did not have any presentation yet, but the first serious discussion about the cost of tokens has started on the documentation level. Looking forward to seeing these presentations!
It's always been this way. America has just been able to coast on being the only remaining major economy after WW2, and exploited the rest of the world instead. That exploitation of the rest of the globe has been mostly optimized now, so those shareholder returns are now coming at the expense of the 90% of Americans who aren't sitting at the table.
I think this is quite reductive. America certainly benefited from being one of 2 major powers after ww2. But unlike the USSR it invested in the world heavily. It rebuilt Allied and Axies manufacturing, and do a lot to revitalize the world economy. They got rich in the process but its not like they did nothing. They invented the internet and cure a ton of diseases. Setup a global order of trade that generates real prosperity.
I guess while I agree that American shareholders do reap incredibly benefits coasting is not really something america does. America is more than just shareholders too. You dont grow the world economy by coasting and you dont make up 25% of the world nominal GDP while only making up ~1/25 its population by coasting its inconsistent with reality.
If you're in a market where a competitor can cut their costs to produce the same quality product, end customers don't know (and if they knew, 95+% of them don't give a crap) how rich you've helped your employees get.
If someone else can make it more efficiently, there's a powerful force for you to also have to improve to match that efficiency.
"Why is the airline experience so much worse than 50 years ago?" "It's massively cheaper per seat-mile, and consumers in aggregate reveal that they prefer the cheapest price that online travel searches, so airlines deliver to that preference."
This is almost entirely on Anthropic and the stupid C suite people trying to push TokenMaxxing. GPT5.5 is much more token efficient, other models are much cheaper, and if used in moderation rather than than trying to get everyone to OpenClaw 24/7 with token leaderboards, it's much more economical.
Also ironically, a lot of GenZ and young Millenials who were already bitter at their employers have used the tokenmaxxing push to sabotage the AI rollouts by burning tokens on stupid shit. It seems to be working.
The other day I had to read a C-suite guy share how he had an epiphany that spending more tokens did not linearly align with more useful features being output by the teams. He was describing it as this breakthrough moment for him, as if it wasn't glaringly obvious that making the KPI "spend more tokens" would result in inefficient token spending, not massive value for the customer.
It's baffling how these people have entirely shut their ears to all the obvious warnings about this, and are now congratulating themselves for their slightly less psychotic outlook and pivoting to blaming the workers for inefficient usage, after specifically forcing them to tokenmaxx.
>It's baffling how these people have entirely shut their ears to all the obvious warnings about this, and are now congratulating themselves for their slightly less psychotic outlook and pivoting to blaming the workers for inefficient usage, after specifically forcing them to tokenmaxx.
It's not baffling. They are a caste, wholly insulated from the consequences of their own actions.
Almost every company is run in a basic dictatorial way. We almost never discuss it, when there is a wide corpus of political Science analysing the pros and cons of governance models that certainly puts it at the bottom.
(Meaning that it's not just business school indoctrination, but a dynamic they've been raised to expect and uphold. Fixing it isn't simply about convincing them of the folly of their approach, because you're attacking their personal sense of self in doing so. Which, I'm to understand, is a no-no, professionally.)
After ejecting anyone who spoke out or were even publicly hesistant against the hard swerve into "just do maximal amounts of AI stuff above all else", they're now surprised to find that everyone that remains is dutifully excited about the emperor's new clothes, and yet he remains mysteriously exposed to the breeze.
If it was so clearly ineffective, why does it get challenged more often and replaced? Existing corporations aren't likely to change, but new startups and work owned coops exist, so why don't they compete?
Maybe ranking it on a scale of best to worse is too simplistic a view, and there are reasons this develops. Maybe it is the best option when there is a good leader, thus such structures dominate, much as a government ran by philosopher kings are better. But this only lasts as long as a wise rule is in charge, and it reverts back to a norm, and eventually, due to pure time and chance, enough bad leaders come on board that slowly dismantle the giants, but this happens at a time scale we don't particularly notice due to how much inertia large corporations can have (before we even get into the less pleasant issues like regulatory capture).
>If it was so clearly ineffective, why does it get challenged more often and replaced?
I supposed you meant "why doesn't it get challenged"?
Well, look at how long it took for a democratic/Republican system to appear and survive. The French 1st Republic was immediately at war with all of Europe (I am not talking of Napoleon at all here, it was before that, when the French King was executed).
Nowadays, good luck getting any kind of financing with an "alternative" governance model. The banks and investors will either refuse or edge by pushing higher return rates on you. The whole system is conservative.
The adage "democracy is the worst system, apart from all others" only becomes true long-term. There are plenty of short-lived democracies back to antiquity, in the middle of the middle ages, during the Renaissance, the XIXth century... All stamped down by "more efficient" dictatorial empires... That aren't here anymore. You can expect the same in the even more cutthroat corporate environment, where fitting the system buys you leverage.
And don't get me stated on startups: most of them seek only an exist strategy. Very few challenge any existing behemoth. They are basically externalized R&D.
One other question I had but wasn't sure if it would leave my previous post too unfocused is "aren't we a bit too early to determine in our current government systems are really the most effective?" This is something that will be decided by political scientists far removed from the current societies who can see how our current societies evolve.
> Almost every company is run in a basic dictatorial way. We almost never discuss it, when there is a wide corpus of political Science analysing the pros and cons of governance models that certainly puts it at the bottom.
Is it not wild that in the Freedom Loving West, we all spend the vast majority of our time as adults living inside tiny totalitarian states?
I think this persists largely because the people atop those tiny states are also the ones behind most of our media apparatus, so they can make it look and feel pretty normal. But that may be a little tinfoil hat of me.
FWIW I don't think it's tinfoil hat at all but when you say things like that on here you get a lot of late-stage-McCarthyists screaming about you being a Communist.
This entire narrative is just made up. Managers know not to reward spending. At best you had some tracking to see who was using it and encouragement for those that aren't to start.
I think this is the part that kills me. This is what many grunts, including myself said from the start. More PRs and more code does not equal value for the customer.
Yes, I think in that way it is dumb. But in another way I think it could be justified as a way to try and blaze some new trails and see what's possible by having users not worry about cost in the beginning.
Sure most token burning ends up being a waste but some ideas pan out?
Not disagreeing but it's another way of looking at it IMO
Yeah, everything is fine until you don’t want to use AI for something because it sucks at that task and then you end up on a PiP because your token burn is low. Why the f*ck are AI Token Use Leaderboards even a thing.
Features that used to take months are now expected in days. Oh you didn’t merge 40 pull requests and deploy to prod 15 times today? Aren’t you using Opus the greatest thing since the invention of the wheel?! What do you mean it’s hard to review 100 merge requests per day? Just have Claude review it! That’s a PiP.
Oh prod is down because people keep deploying code that nobody even freakin’ read? Just have Claude fix it! What do you mean it’s doesn’t work well? Just burn more tokens or you’re on a PiP.
Surely there wouldn’t be malicious compliance by people that would prefer to use the right tool for the job instead of having this crap shoved down our throats by management by threat of termination.
Does this happen? I’ve never been at a company that measures employee performance by token burn targets. I suspect most companies don’t do that, but I could be wrong obviously.
I'm with you that people are insanely hyped about Claude Code in particular when e.g. Codex isn't far behind (and with recent models I actually prefer it).
But I'm going to need a citation for this:
> a lot of GenZ and young Millenials who were already bitter at their employers have used the tokenmaxxing push to sabotage the AI
The 3 people on reddit doing this don't even register on a company budget. What seems more plausible to me is that budgets were calibrated to spending before agents were actually useful, and late '25/early '26 changed the pattern significantly.
Codex is actually significantly better than Claude Code now, assuming you have a clear idea of what you want to do and how. Claude's secret sauce is that it'll run off and do stuff that's mostly right without a lot of prompting, but that also makes it willful/disobedient and causes it to be bad for "finishing" work, since it'll circle around your objective in an opinionated way.
> assuming you have a clear idea of what you want to do and how
I mean, if I have a sufficiently clear idea of what and how, then surely just coding it manually would work significantly better. Unless maybe I am a painfully slow typer.
Without some level of "actually I'm not sure exactly" permitted, then I'm not really sure what LLMs bring to the table.
If I want to create a web app with a back-end, database, and some services, and I tell codex to do that with a specific stack and using specific paradigms to keep the code performant and maintainable, it's still a win over coding it by hand, as models can emit ~200char/sec compared with maybe ~10 for a really fast human. There's up front planning cost, and you will have to go back and massage some of the outputs a little bit if you're particular, but for sizeable tasks it still comes out to be a big win.
If you're just working on a single react component or an algorithm to do stuff with data, there's less chance to amortize the up front planning and verification so it comes out more of a wash.
Even when you have a clear idea of what you want, there are still hundreds of decisions you need to make while building it, both big and small. Everything from what to name your database tables and columns to what data structures are optimal and what the API payloads should look like and what the tech stack should be. Anyone with a sufficient level of experience in this field has made these types of decisions dozens of times and at some point it becomes more practical to have an AI do it for you and for you to quickly skim it.
For example I want to make it so that users receive an email when their password is changed. I can either do it myself, which requires reviewing and remembering code I’ve written five plus years ago and then wiring everything up and obsessing over the wording of the email. Or I can give a two sentence instruction to the AI, work on something more meaningful while it is doing its thing, and then test it in under 60 seconds when it is done.
Counterpoint: based on a lot of anecdotes here, the most likely people to burn tokens aren’t GenZ but managers using ChatGPT to respond to questions or otherwise as an outsourcing of their job. There aren’t enough GenZ in the workforce to back your claim in my opinion.
Token efficiency where instead of the AI burning money at 1:3 instead of 1:5 isn’t quite a winning argument.
There's no way managers using LLMs to answer emails are burning tokens at a comparable rate to someone trying to utilize inference in production systems is.
Excessive token burning as a tactic to annoy your employer probably does the opposite - it probably makes your employer money.
The more tokens you burn, the more demand for hardware there will be. More demand for hardware means higher stock prices -> more money in your employer's pocket.
The only question is how long that can last. If taken to an extreme, the output of the AI will get worse over time, and if it gets bad enough, for long enough, people will use it less and less, and demand will slowly evaporate.
My point is NOT: I hope this all comes crashing down.
My point IS: tokenmaxing is bad. AND weaponizing it will not have the intended effect, but in fact, it will, in the short term, do the opposite.
> The more tokens you burn, the more demand for hardware there will be. More demand for hardware means higher stock prices -> more money in your employer's pocket.
If you work for NVIDIA, sure, but otherwise, this makes no sense to me.
I'm guessing you aren't invested in any of these stocks. I'm also guessing, to take the perspective of a tokenamxxing GenZ employee, that their bosses are.
Right, because every $100 spent on tokens results in more than $100 of profit from NVIDIA stock. That totally makes sense. But why so indirect, why not buy NVIDIA GPU and trash them? That should increase profit even more, since you're cutting out the middlemen.
All that aside, refusal to participate in bullshit doesn't imply some "plan" to bring the company down, it's really just refusing bullshit. It's not "5D bullshit", i.e. a move in the same bullshit game they are refusing
And if you offer someone 5 bucks to help burn down their own house, that of all their friends, so someone can make 5000 bucks, then they might just refuse, not because they have some plan, but precisely because refusal or letting themselves get hollowed out as well is the only option they're given. It's not like any of the hollowed out people listen to them in earnest.
Young people, which is crazy to me, often want some kind of stake and participation in the world rather than being disposable slaves, they want to do work that makes sense, that enables them and others to grow and prosper, in order to prepare them to become stewards of that world later on, which they want to make better and not just exploit, generally. Mostly, some of them of course are assholes or stupid, but in general, people come to this world way less fucked up than the world wants them to be.
Overheard recently: "Thanks to AI we're producing more code and more MRs, faster than ever, but the milestones aren't getting hit any sooner. Actually the opposite, if anything."
I wonder how widespread that phenomenon is. Perhaps it's no wonder the prominent actors are trying to rush to IPO...
Goodharts law. The metrics were always measuring the wrong thing, and now that we've finally optimized for the wrong thing successfully management will be forced to admit it and move on to another, slightly different, metric that doesn't actually equate to shareholder value.
It doesn't matter what the line actually measures, just that it goes up.
That's exactly what's happening. Many claim they are more productive with AI, but individual rise in productivity just doesn't translate to projects being completed any sooner.
And by "projects", I mean corporate ones with big teams involved. Hobby projects actually do get finished much faster.
In the corporate world, writing code was always a small chunk anyway. Iirc something like 30% of employee time. Getting 50% faster there still only gains back 10% of your time.
To further complicate matters, you had to spend on AI software and potentially additional on legal/risk/security/compliance to enable that.
The smaller the company, the bigger that % of coding is of total time (all the way down to hobby where the majority of time is spent coding)
My take from working at a Big Corp is that individuals using coding agents can increase velocity substantially and produce good quality work, assuming they are proficient with the tools. But it falls apart quickly when you have a team trying to work together.
imo we either need to centralize the agent and and submit plan, spec, reference doc MRs rather than submitting code changes. Or develop SCM systems/workflows that incorporate plan/spec/reference/prompt metadata with code so intent can be factored into merges.
We're planning for Q1 and don't have requirements yet for features, but are expected to have detailed estimates ready... This has happened before AI, this is happening after AI. The problem has rarely been "it takes too long to build
". The problem was "what do you want me to build? wait, no, not like that. now we must iterate"
I wish I could do the same thing. Coworkers would be allowed to ask me X number of questions per month, and once they hit that limit I get the rest of the month off.
Just a week ago, Anthropic barely breaking even was hailed as AI companies being close to profitability much earlier than forecast.
In fact it is all smoke and mirrors, pure mania from C-level executives out of their depth trying to one-up each other with company money, and they aren't even close.
I didn't look into the numbers myself, but Ed Zitron claimed in his podcast that the only reason Anthropic looks sort of profitable was due to some rather clever accounting (not illegal) and a spending that didn't happen, I think. Things haven't actually improved much, but if you zoom into a very selective part of the fiscal year it looks promissing.
I really can't tell what is going on with AI these days. I hear AI labs claiming theyre profitable or close to it. I hear companies say they're dubious the juice is worth the squeeze. I've seen anecdotal claims of a measurable increase in productivity of 2x in PRs created and merged coming in at cost 20% of engineering employee budget. Others say they're still getting no value (which I doubt). Simon Willison's recent post went into debunking the AI sticker shock claims somewhat. Either way this seesawing between new golden era and the greatest VC money furnace is becoming exhausting.
I'd like to see real numbers at this point, and this article is just a few bullet points that link to other articles. Talk is far cheaper than tokens and I'd like to have a workflow that I can rely on being there in six months.
AI Has empowered people to build things much more quickly. Not slop if you are even a little conscientious about how you use it. What it does not do fix the human structural problems. If you are solving the wrong problems you aren’t doing anything useful. Just because you can now take an idea to near completion doesn’t mean it was worth doing, but now you spent tokens and a lot of your mental bandwidth to finish it. Or worse you let it become slop and it will fall apart if you even look at it funny.
Previously if I needed to automate something I thought really carefully about it. Now, I still think really carefully about it. I had fun AI coding some tools I always wanted but they were just pet projects for me. I had fun AI slop coding a couple of things, but it was not good software. But if you have a clear and valuable target? AI can absolutely get you there.
Multiply that across all your colleagues and a lot /seems/ to be happening, but what is actually moving the needle?
>Corporate leaders are starting to question whether soaring AI spending is delivering meaningful returns.
This isn't surprising. Ive recently run into quite a few rabbit holes where AI is bad enough that its much more efficient to do it myself. I wanted to refactor some code, gave it a design pattern to go towards, some specific classes and methods, etc. making it a well described problem. AI just couldn't do it satisfactorily. The code was ugly, overly verbose, and after multiple tries with multiple prompts saying to keep things simple. They still would introduce new classes, useless fields, etc.
This was GPT 5.5 and codex. The specific model and harness isn't that important here. AI could do it. But the issue seems to be that there are some tasks where AI kind of falls over and provides poor results. It was easier, better, and faster for me to just do it myself. I have found a lot of cases where AI is great. If you have a UML diagram already, or translating code from language x to y, fixing unit test failures, generating boiler plate. But I can definitely see if people are using large amounts of AI for writing code, analyzing code, etc. that they are not actually seeing returns.
"An AI consultant tells Axios one of their clients recently spent half a billion dollars in a single month after failing to put usage limits on Claude licenses for employees."
Like physically, how could this even happen?
I imagine some scenarios where you're throwing a big data set at an LLM could add up quickly.
"Check all of Confluence for outdated and conflicting info"
"Review all the legal contracts over the past 10 years"
"Evaluate the source code in all our dependencies for vulnerabilities"
All of those would be fairly straight forward for a single person to kick off without thinking about the data set size. Especially if they have Claude set to Opus 4.7 x high effort for everything.
Even someone saying something like "rewrite the 25 year old Java monolith in rust" as a PoC and leaving it running in the background for a week
Let's not forget about the environmental impacts. It's crazy that people are willfully burning so much energy for almost no return. And we thought bitcoin was bad... This is just completely irresponsible, if not sociopathic behavior.
Bitcoin doesn't have long conversations with you, including deeply technical ones, in a way that you thought only a human being could just 5 years ago.
They are not the same.
I agree fully with you on the potential for energy waste. We always do that, though, with nearly everything. How many of today's jet plane flights really needed to happen? The question is how much value people feel they're getting. People are having a whole lot more feelings about AI than they ever could about cryptocurrencies, and that train aint stoppin'.
How much longer before we get to the “I get cancer. I kill Jack” stage? Isn’t uncontrolled growth a sign of cancer? Or is it more virus like where it just continues to grow without ever reaching a balance with its surroundings?
This is because the current AI approach relies on AI to be a glorified search engine – know everything about everything requiring enormous, ever growing models, and demanding search-engine like near instant responses requiring bigger more complex chips and sprawling data centers to run them in. This leads to a loop demanding ever bigger models, updated at a more and more expensive cost, and chipsets that become much more expensive to deploy.
If you move those things to software and utilize tools that are cheap at scale (databases, web search etc.) the hardware arms race ends and the price becomes sustainable. With the right tools preparing dynamic context for a conversation, models are used for their reasoning and not for their knowledge. And waiting even a minute or two for a model to prepare a response, evaluate it, and iterate to improve quality makes a huge difference.
Corporation always keep doing this, they hire lousy programmers then they try to make them produce good code by forcing them to use all kinds of hyped cutting edge technologies.
I'm sure there is an explanation for why they keep doing it.
> "Most people default to automating tasks they dislike rather than tasks most valuable to the company," Sophia Velastegui, CEO of Velastegui Ventures and former chief AI officer at Microsoft, told Axios. Instead, they should focus on using AI to drive revenue.
That's right! Work, slave, pain away every day! How dare you make your life a bit less miserable?!
"“It’s a real dollar investment,” says Tan, speaking for Claudeholics who go full blast. “You actually have to spend six to seven figures on tokens—I’m on a run rate to do seven figures this year.”"
Has Wired gone evil now too? Just because some SV programmer circles who excel at marketing switched from Ayahuasca to a new drug does not mean everyone is going to use the new drug.
There are literally no good programmers mentioned in that article. They are all boiler room VC types. Shame on Levy, this article is planted or has has no clue.
The AI fever pitch has done a great job at exposing which companies were run with a degree of sanity, versus who bought blindly into the hype train narrative of worker replacement and went all-in.
Look, LLMs thrive when they’re given structured data that’s well annotated, clear direction, and treated as the probabilistic machines they are. Not one of those meshes with the AI narrative of “works on existing stuff, requires minimal guidance, and can behave deterministically.”
I said as much in 2024 when my employer at the time was grading folks on AI usage while my role was entirely deterministic in nature. It didn’t resonate with specific leadership then, it doesn’t seem to be doing so in the larger market now, and unfortunately not one of these dolts will suffer any consequences for their organizational myopia.
I mean most managers in my incredibly subjective experience are indistinguishable from the probability slot machines that are LLMs, and those who weren’t were often managers I’d walk through fire for.
I’d love to see more research on the efficacy of LLMs as organizational middle management, but I fret that without sufficient anonymity protections for staff they’ll just do the same biased shit humans do.
Just wait until companies are dependent on on it. When their employees can't think without it. When their AI generated codebase is such a mess they'd need a rewrite to understand it without AI. When they've got AI embedded in all their internal processes and tools. Then massive price hikes will come because they've been bent over a barrel and they'll have no alternative that isn't at least as painful in the short-term as letting the AI company fuck them. The long term won't matter then because any company capable of seeing past the short term wouldn't let themselves get into that position in the first place.
> Corporate leaders are starting to question whether soaring AI spending is delivering meaningful returns.
This is act one the AI bear market. Yes I know everyone screams “bubble”. Let me explain the scenario I have in mind.
1. AI booms because the technology seems to actually have promise of revolutionizing how work gets done. It can do your taxes! It can drive Excel! It can act as a CEO! It can code up full apps and SaaS products! It can replace this vendor or that! You know the drill.
2. Every company must in corporate AI or be seen as obsolete. Having a bad quarter? Announce that you are “seeking to explore opportunities to develop an AI integration plan framework” for your plumbing business. Massive AI compute buying happens. While two of the three major AI houses are not publicly traded proxies like Nvidia and RAM manufacturers are so the market rips higher and higher. Nvidia trades as if it is already 10+ years from now, every company out there has adopted AI perfectly, and it is delivering huge profits to them.
3. Reality checks start pouring in. Turns out that not only is AI expensive (a problem that presumably will be taken care of with time and development), but that the technology itself just isn’t suitable for everything. (IMHO it’s great at augmenting a power user but it is terrible at interacting directly with customers). We start seeing individual companies change tone on investment. They can’t stop it due to momentum but they are starting to shift the narrative to warn of what comes next. This is where we are.
4. Numbers come in. Earnings show what the actual ROI is. Some companies do benefit, but crucially we see examples of where investing in AI destroys value. I think this happens when replace crucial parts of their workforce with agents and find that they lost in-house expertise, when customers left due to worse products, or simply when AI was roughly as expensive as human labor without being significantly more productive.
5. The market stumbles. What do you mean AI won’t take over every corporate America?! Surely that can’t be right! Nvidia and other proxies flag.
6. In a late to the game rush Anthropic and OpenAI IPO fearing that the market has noticed that the emperor has no clothes. Their internal numbers turn out to be scary: very high revenue but no path to insane profitability. They quickly get included in QQQ and maybe even S&P500 but as their IPO price is the highest they trade they drag the broad market indexes down. This is leveraged by the Nvidia proxy status.
7. Infrastructure course correction. Hyperscalers who started huge datacenter buildouts cannot justify it. They pay contract penalties and get out of some of the projects, writing down losses. The market fully melts.
—-
I think there is a competing market downturn fueled by the affordability squeeze. Basically while AI spend is corporate driven, the biggest investors are consumer companies. Of the hyperscalers you can maybe argue that Microsoft is not a B2C fully but it is close. If consumers don’t have money to spend, hyperscalers take a hit, investment slows from there, and AI is hit directly by it.
I think either scenario is likely, it’s just which happens first. But right now the market is sprinting down a tight rope and trading like that tight rope has no end and that the sprinter never makes a mistake regardless of wind changes. Everything has to go right for a very long time to justify valuations. One stumble can stop it all.
> Instead, they should focus on using AI to drive revenue.
There is a complete disconnect between wages of employees and company's revenue => Why aren't employees working towards revenue? What a mystery. Children, let's help Elmo solve this mystery.
And then random mass layoffs to make numbers for shareholders look great in quarterly reports. Surely this motivates people work to their fullest potential and to care for company's revenue.
I found this claim interesting so I looked into it. Everything I can find shows that the intuition is accurate.
Companies with EOSP programs outperform those that do not in the market by about 17%.[0] Companies that perform layoffs, despite short-term stock boosts, underperform on a period of years showing a 14% decline in their Return on Assets (ROA) in the years following the layoffs.[1]
Even better there’s a complete disconnect between revenues and metrics we use to measure productivity. Corporate wants to believe there’s numbers you can use to measure knowledge workers like widget makers where there’s really not much that’s effective beyond revenue.
There's a very real possibility that we end up with a bunch of companies simultaneously realizing that they just don't have any good ideas to leverage AI on and rushing more code out the door doesn't lead to tangible value. Many execs are going to realize that they're no longer growth companies and that paying $500k a year for engineers no longer makes sense...
If that's the case, then I do expect the AI bubble is going to pop spectacularly next year as token budgets are going to collapse. The damage to the tech industry is going to be catastrophic. If you think the job market is bad now, wait until data center spending goes off a cliff.
If I understood correctly, a few months ago Anthropic and OpenAI both started charging per-token billing at API pricing for Enterprise customers? i.e. representing roughly a 10x price increase? That's kinda nuts.
Well, if AI has a massive sticker shock attributed, so we should target the high value roles and should save money, right?
So im looking at CEO, CTO, CFO, and all the chief-something-officer. If LLMs are that totally amazing at thinking, then we should be targeting upper management, not the workers.
That would save a LOT of money for the shareholders! /snark
So it goes. Wage theft dwarfs the amount lost to street-level theft, robbery, burglary, etc., combined. The economic stimulus from correcting even a portion of annual wage theft would represent complete coverage of those violent thefts - economically-speaking, there would be no reason for criminals to carry them out. Why rob a gas station to get your drug money? Everyone around you is making enough extra at work that bumming a dollar here and there covers it. That sort of thing.
But good forbid we actually correct a major social ill at the expense of the people who profit from it.
Exactly. And even worse is that crime itself is firmly rooted in politics.
Example: You work at walmart, as a cashier. You drop a $100 bill on the floor and pocket it from the till. You're caught, cops called, and you are arrested.
Example: You work as a manager at the same walmart. Store manager says labor costs are too damned high. So you go in on 10 employees, and edit their timecards to cut $100 from each of them, totaling $1000. IF you are caught, police will not respond. Instead, it is a "civil matter".
This is a bit dated chart, but its still very much correct. https://www.tcworkerscenter.org/2018/09/wage-theft-vs-other-... But notice the wage theft types are all "civil matters", and the non-eage theft are heavily criminally prosecuted. And who does those? Predominantly poorer people.
We also see criminality differences between charges of freebase cocaine versus crack cocaine. Crack was what black people smoked, so sentencing was like 10x of freebase.
When you start looking at how laws are apllied, its almost always the same pattern: those at the top are a civil matter. Those who report to the top are a criminal matter. Those at the bottom are "charged with the fullest maximum punishment".
Right. Though I want to focus on a specific aspect of your hypothetical (which is essentially real):
If the manager hadn't stolen $100 from the cashier, there would have been a MUCH weaker incentive for the cashier to steal themselves.
This is the crux around which everything else turns: we are effectively post-scarcity, as far as production is concerned. As a society, we purposely create theft, and debt, and the associated desperation and crime, as a matter of policy. As a choice.
If you were eradicate wage theft, you would essentially eradicate the internal logic of street theft, in the vast majority of cases. We could just... not have theft. But by not prosecuting wage theft, we, as a society, have decided that we condone and even need theft.
CFO: “Um, this is costing a ton and we’re not seeing savings or efficiency materialize.”
CEO: “Are we getting any value out of this?”
COO: “Not really, and frankly I’m getting annoyed at all the AI slop turning up all over.”
CEO: “OK, well, let’s do a big layoff and then I’ll just say it was because of AI. Hopefully folks won’t blame me for the mess and I’ll just talk about how amazing AI is.”
Lesson #1 from business school : take all credits, put the blame on others, if there no easy scapegoat blame the "economical context"
Lesson #2 and beyond : just see lesson #1, it's enough. You've made it and it's ALL thanks to this amazing business school degree you got, now go profits with your new already wealthy peers.
I have a literal ledger tracking the number of times executives have used the phrase "evolving capital markets"
It's amazing to me how subtle the difference can be between a business leader that seems to know what they're doing and one that clearly does not. Same words, even, but from one mouth it's compressing a complex thought and from another it's word salad to give the appearance of complex thought.
These are golden times for AI consulting. We're getting into that business because it is up for grabs. Everybody as AI FOMO right now and begging people to take their money. Exactly what you are highlighting. And, yet, nobody has the beginning of a clue what to do or how to do stuff with AI. That's classic consulting territory.
My analysis of the whole space is that all the tech people are focused on completely the wrong end of the problem space (model quality, training, token cost, etc.). They are geeking out on that stuff. Worse, they are mostly not even that good at using their own tools. I regularly talk to so called "AI native" companies that are hiring enterprise sales people to do sales like it's 2006 instead of 2026. Their product is AI native, but their companies aren't.
The real issues are on the low level plumbing end. How do you connect all the data silos you have. Are they even the right silos still? And once you fix that: how do you get organized around tapping into those. What guardrails do you need? How do you deal with compliance issues? Etc. You don't need to do a lot to get value back. Most people haven't gone there yet.
Mostly people actually already have and pay for the tools that would enable a lot more if only they knew how: ChatGPT, Codex, Manus, Perlexity Computer, Claude CoWork, etc.
What you are outlining is funny but true. The issue isn't money or tools but figuring out what to do with the tools you already bought.
>Corporate leaders are starting to question whether soaring AI spending is delivering meaningful returns.
We should start to question whether soaring CEO salary spending is delivering meaningful results.
Ah, that's not how it works : CEOs are rich so they are successful. They are successful so they are rich. Q.E.D.
> CEOs are rich so they are successful.
"Successful" in which sense?
The most reasonable definition "Successful in convincing other people to give them money" is rather a tautology. :-)
A very related problem concerning your argument is well-known for decades: the Principal–agent problem:
> https://en.wikipedia.org/wiki/Principal%E2%80%93agent_proble...
It says that the self-interest of an agent (e.g. the CEO) is not identical to the interests of the principal (e.g. the shareholders).
Why do I mention this? The CEO being rich is a measurement of the financial success of the agent, but not of the principal. You are rather interested in measuring the success of the principal.
As you might see in other comments the tautology is the joke.
Maybe in your fantasy world.
In reality people are rarely rich based on merit alone.
EDIT: Sorry it was a really clever joke.
It's sarcasm, obviously.
>>>EDIT: Sorry it was a really clever joke.
I sincerely hope that was your attempt at sarcasm.
I think you missed the joke.
About 2 decades ago, there were multiple studies showing that C-level execs who offshored work/production got 18% higher compensation. I'm sure that the same bizjournals are claiming that C-level execs that "embrace" AI are getting similar compensation premiums over anyone who takes a "wait and see" approach. The quantity of AI appearing in bizjournals reminds me of Paul Graham's essay on The Submarine.
https://www.paulgraham.com/submarine.html
All we can really do is point and laugh. Boards don't listen to workers, and I bet most boards will be okay with a little spending oopsie-daisy because it was 'try shit and see if it works out'
We are since at least the 2000s when the world wide web revolution finally made it possible for Google, Apple, Amazon etc... to become the behemoths they are. And maybe even before that.
And the answer is no, because when chief officers succeed it's because of their genius and forward thinking posture, but when they fail it's none of their fault, so they are shielded in an echo chamber that produces such delusions and psychosis.
Way before that, as usual we can attribute quite a lot of stupidity in corporate governance to Jack Welch. Execs really bought into Welch's schtick wholly, they went to MBA schools praising Welch's management style, read his books, or at least got taught by people who had bought wholly into it.
So much time has passed that I believe truly the current crop of execs don't know any better, they think this status quo is the only way to manage companies. They aren't really wrong since the incentives are there, and they continue to reap rewards from doing it.
Agreed. For a long time Jack Welch was lauded as "The Greatest CEO in History" for burning GE to the ground in a way that made a lot of money (while destroying a company that took 100 years to build and employed 400,000 people).
By the time he died (pretty recently actually, 2020!), it was pretty obvious what kind of legacy he was leaving behind. Which is probably why his family was very careful to keep his burial location secret, presumably to keep people from peeing on his grave.
what a bizarre non sequitur
This is an important inflection point.
When tokens get correctly priced, all of the insane over-investment in capital will need to draw back: buying data centers, semiconductors, and politicians.
Even then, it won't be right-priced with regard to actual costs. The environmental impact should have been priced in from the beginning. There seems to be a parallel with subsidizing fossil fuels, under pricing them which encourages over dependence, ignoring the real costs society will pay later.
It rather looks like chatgpt/antrophic enterprise tokens and API calls are too expensive. Competition is quite strong on openrouter.
However, the real problem is running wild with token burning. With parallel agents calling subagents you can burn lots of tokens per minute. Especially with thousands of engineers.
I've yet to see any compelling data about inference being particularly expensive. For local LLM models, that are becoming increasingly viable, it's dirt cheap. The same is also true in image gen world where now even a heavily dated GPU can cheaply and quickly produce high quality images.
I also think the image gen world is a useful analog because there are a million sites, presumably still making money, with markups that are multiple orders of magnitude off their costs. They're feeding off user ignorance that was, at least in part, artificially seeded by implying high costs for image gen back in its day. Though it's possible/probable that the initial training runs were expensive, but that's a one-and-done cost.
We use GH Copilot at work and this week sat for a presentation by GH about optimizing token usage and maximizing ROI on tokens used. Anyone else get this presentation? They didn’t have time for questions because they had to run and give it to the next big enterprise on their list…
They basically said that everything is too expensive, you have to watch it like a hawk. It was as if they poured a bucket of cold water on the room. People were wondering how they could do anything faster with all these strategies. And then “sorry no questions. Bye!”
Interesting. We use Kiro here and looking at the public pricing subscriptions and it's benefit to my workflow, it is clearly a significant productivity increase per dollar spent. And we were told we have a signed a deal that is better than that public pricing. They recently just enabled overages on everyone's account so that people aren't throttled and they are shifting people up/down tiers as required behind the scenes to align with their actual usage.
However when the 'cost' to do something is relatively flat the cost/benefit analysis is going to depend on the value of the person being enabled. Someone making $60k a year using AI to gain a 20% output improvement may not be worth the cost but someone making $160k a year would.
Do you use Kiro specs? I have been loving using Kiro for the same reasons, and and just starting to use the spec feature.
Did you have any good tips for optimizing usage?
The ones I’ve stumbled upon seem to be: switch models based on task complexity, use tooling like ASTs and compression, disable unused MCPs, compact often, be verbose with input to give clear guidance…
I don't think compacting often is good for saving money. It generates more output tokens and then the input is no longer from cache, which is priced differently...typically very differently.
We did not have any presentation yet, but the first serious discussion about the cost of tokens has started on the documentation level. Looking forward to seeing these presentations!
> They didn’t have time for questions because they had to run and give it to the next big enterprise on their list…
This is such bullshit. Surely they could have recorded the shared part of the presentation and then spend all their time answering the questions?
Just hire people and pay then a fair wage and let everyone get richer together ffs
When did American capitalism become such a zero sum game
When the powers that be decided labor didn't matter and the only thing that mattered was capital.
One will never get rich on wages. The only way to get rich is through asset manipulation and rent-seeking.
It's always been this way. America has just been able to coast on being the only remaining major economy after WW2, and exploited the rest of the world instead. That exploitation of the rest of the globe has been mostly optimized now, so those shareholder returns are now coming at the expense of the 90% of Americans who aren't sitting at the table.
I think this is quite reductive. America certainly benefited from being one of 2 major powers after ww2. But unlike the USSR it invested in the world heavily. It rebuilt Allied and Axies manufacturing, and do a lot to revitalize the world economy. They got rich in the process but its not like they did nothing. They invented the internet and cure a ton of diseases. Setup a global order of trade that generates real prosperity.
I guess while I agree that American shareholders do reap incredibly benefits coasting is not really something america does. America is more than just shareholders too. You dont grow the world economy by coasting and you dont make up 25% of the world nominal GDP while only making up ~1/25 its population by coasting its inconsistent with reality.
If you're in a market where a competitor can cut their costs to produce the same quality product, end customers don't know (and if they knew, 95+% of them don't give a crap) how rich you've helped your employees get.
If someone else can make it more efficiently, there's a powerful force for you to also have to improve to match that efficiency.
"Why is the airline experience so much worse than 50 years ago?" "It's massively cheaper per seat-mile, and consumers in aggregate reveal that they prefer the cheapest price that online travel searches, so airlines deliver to that preference."
Can you think of a point in history when it wasn't.
Despite all of the problems that exist were still the best off.
This is almost entirely on Anthropic and the stupid C suite people trying to push TokenMaxxing. GPT5.5 is much more token efficient, other models are much cheaper, and if used in moderation rather than than trying to get everyone to OpenClaw 24/7 with token leaderboards, it's much more economical.
Also ironically, a lot of GenZ and young Millenials who were already bitter at their employers have used the tokenmaxxing push to sabotage the AI rollouts by burning tokens on stupid shit. It seems to be working.
Are you saying that making a leaderboard of who is spending the most is going to be expensive?
They couldn't see that coming, but for sure they can predict how the future will be when it's time to sell their "visions" of the world.
Meanwhile, sheep's are going to believe and max their token usage with their own wallet. "You are so be left behind if you're not".
It's a mass psychosis. The only winners here are the hardware manufacturers, like nvidia for instance.
The other day I had to read a C-suite guy share how he had an epiphany that spending more tokens did not linearly align with more useful features being output by the teams. He was describing it as this breakthrough moment for him, as if it wasn't glaringly obvious that making the KPI "spend more tokens" would result in inefficient token spending, not massive value for the customer.
It's baffling how these people have entirely shut their ears to all the obvious warnings about this, and are now congratulating themselves for their slightly less psychotic outlook and pivoting to blaming the workers for inefficient usage, after specifically forcing them to tokenmaxx.
>It's baffling how these people have entirely shut their ears to all the obvious warnings about this, and are now congratulating themselves for their slightly less psychotic outlook and pivoting to blaming the workers for inefficient usage, after specifically forcing them to tokenmaxx.
It's not baffling. They are a caste, wholly insulated from the consequences of their own actions.
Almost every company is run in a basic dictatorial way. We almost never discuss it, when there is a wide corpus of political Science analysing the pros and cons of governance models that certainly puts it at the bottom.
>They are a caste
Sometimes literally.
(Meaning that it's not just business school indoctrination, but a dynamic they've been raised to expect and uphold. Fixing it isn't simply about convincing them of the folly of their approach, because you're attacking their personal sense of self in doing so. Which, I'm to understand, is a no-no, professionally.)
I should have perhaps said "galling" instead.
After ejecting anyone who spoke out or were even publicly hesistant against the hard swerve into "just do maximal amounts of AI stuff above all else", they're now surprised to find that everyone that remains is dutifully excited about the emperor's new clothes, and yet he remains mysteriously exposed to the breeze.
If it was so clearly ineffective, why does it get challenged more often and replaced? Existing corporations aren't likely to change, but new startups and work owned coops exist, so why don't they compete?
Maybe ranking it on a scale of best to worse is too simplistic a view, and there are reasons this develops. Maybe it is the best option when there is a good leader, thus such structures dominate, much as a government ran by philosopher kings are better. But this only lasts as long as a wise rule is in charge, and it reverts back to a norm, and eventually, due to pure time and chance, enough bad leaders come on board that slowly dismantle the giants, but this happens at a time scale we don't particularly notice due to how much inertia large corporations can have (before we even get into the less pleasant issues like regulatory capture).
>If it was so clearly ineffective, why does it get challenged more often and replaced?
I supposed you meant "why doesn't it get challenged"?
Well, look at how long it took for a democratic/Republican system to appear and survive. The French 1st Republic was immediately at war with all of Europe (I am not talking of Napoleon at all here, it was before that, when the French King was executed).
Nowadays, good luck getting any kind of financing with an "alternative" governance model. The banks and investors will either refuse or edge by pushing higher return rates on you. The whole system is conservative.
The adage "democracy is the worst system, apart from all others" only becomes true long-term. There are plenty of short-lived democracies back to antiquity, in the middle of the middle ages, during the Renaissance, the XIXth century... All stamped down by "more efficient" dictatorial empires... That aren't here anymore. You can expect the same in the even more cutthroat corporate environment, where fitting the system buys you leverage.
And don't get me stated on startups: most of them seek only an exist strategy. Very few challenge any existing behemoth. They are basically externalized R&D.
Apologies on the typo.
One other question I had but wasn't sure if it would leave my previous post too unfocused is "aren't we a bit too early to determine in our current government systems are really the most effective?" This is something that will be decided by political scientists far removed from the current societies who can see how our current societies evolve.
> Almost every company is run in a basic dictatorial way. We almost never discuss it, when there is a wide corpus of political Science analysing the pros and cons of governance models that certainly puts it at the bottom.
Is it not wild that in the Freedom Loving West, we all spend the vast majority of our time as adults living inside tiny totalitarian states?
I think this persists largely because the people atop those tiny states are also the ones behind most of our media apparatus, so they can make it look and feel pretty normal. But that may be a little tinfoil hat of me.
Is it too late to scratch your final sentence?
FWIW I don't think it's tinfoil hat at all but when you say things like that on here you get a lot of late-stage-McCarthyists screaming about you being a Communist.
Hmm. That hasn't been my experience of HN, but if so, all the more reason to say what you really think. /$.02
This entire narrative is just made up. Managers know not to reward spending. At best you had some tracking to see who was using it and encouragement for those that aren't to start.
I think this is the part that kills me. This is what many grunts, including myself said from the start. More PRs and more code does not equal value for the customer.
You get what you measure.
Yes, I think in that way it is dumb. But in another way I think it could be justified as a way to try and blaze some new trails and see what's possible by having users not worry about cost in the beginning.
Sure most token burning ends up being a waste but some ideas pan out?
Not disagreeing but it's another way of looking at it IMO
I’m an “old millenial” and the excessive burning of tokens will continue until working conditions improve.
Working conditions are fine, I simply am not incentivized to be efficient with tokens.
Yeah, everything is fine until you don’t want to use AI for something because it sucks at that task and then you end up on a PiP because your token burn is low. Why the f*ck are AI Token Use Leaderboards even a thing.
Features that used to take months are now expected in days. Oh you didn’t merge 40 pull requests and deploy to prod 15 times today? Aren’t you using Opus the greatest thing since the invention of the wheel?! What do you mean it’s hard to review 100 merge requests per day? Just have Claude review it! That’s a PiP.
Oh prod is down because people keep deploying code that nobody even freakin’ read? Just have Claude fix it! What do you mean it’s doesn’t work well? Just burn more tokens or you’re on a PiP.
Surely there wouldn’t be malicious compliance by people that would prefer to use the right tool for the job instead of having this crap shoved down our throats by management by threat of termination.
> on a PiP because your token burn is low
Does this happen? I’ve never been at a company that measures employee performance by token burn targets. I suspect most companies don’t do that, but I could be wrong obviously.
I know people that were laid off because they were low on the “token leaderboard” so yes it happens.
I doubt it, but I've seen dumber stuff like laying off people because they think AI will replace them
I'm with you that people are insanely hyped about Claude Code in particular when e.g. Codex isn't far behind (and with recent models I actually prefer it).
But I'm going to need a citation for this:
> a lot of GenZ and young Millenials who were already bitter at their employers have used the tokenmaxxing push to sabotage the AI
The 3 people on reddit doing this don't even register on a company budget. What seems more plausible to me is that budgets were calibrated to spending before agents were actually useful, and late '25/early '26 changed the pattern significantly.
Codex is actually significantly better than Claude Code now, assuming you have a clear idea of what you want to do and how. Claude's secret sauce is that it'll run off and do stuff that's mostly right without a lot of prompting, but that also makes it willful/disobedient and causes it to be bad for "finishing" work, since it'll circle around your objective in an opinionated way.
https://finance.yahoo.com/sectors/technology/articles/nearly...
Hey, would you mind elaborating a bit on this:
> assuming you have a clear idea of what you want to do and how
I mean, if I have a sufficiently clear idea of what and how, then surely just coding it manually would work significantly better. Unless maybe I am a painfully slow typer.
Without some level of "actually I'm not sure exactly" permitted, then I'm not really sure what LLMs bring to the table.
If I want to create a web app with a back-end, database, and some services, and I tell codex to do that with a specific stack and using specific paradigms to keep the code performant and maintainable, it's still a win over coding it by hand, as models can emit ~200char/sec compared with maybe ~10 for a really fast human. There's up front planning cost, and you will have to go back and massage some of the outputs a little bit if you're particular, but for sizeable tasks it still comes out to be a big win.
If you're just working on a single react component or an algorithm to do stuff with data, there's less chance to amortize the up front planning and verification so it comes out more of a wash.
Even when you have a clear idea of what you want, there are still hundreds of decisions you need to make while building it, both big and small. Everything from what to name your database tables and columns to what data structures are optimal and what the API payloads should look like and what the tech stack should be. Anyone with a sufficient level of experience in this field has made these types of decisions dozens of times and at some point it becomes more practical to have an AI do it for you and for you to quickly skim it.
For example I want to make it so that users receive an email when their password is changed. I can either do it myself, which requires reviewing and remembering code I’ve written five plus years ago and then wiring everything up and obsessing over the wording of the email. Or I can give a two sentence instruction to the AI, work on something more meaningful while it is doing its thing, and then test it in under 60 seconds when it is done.
Citation needed.
Counterpoint: based on a lot of anecdotes here, the most likely people to burn tokens aren’t GenZ but managers using ChatGPT to respond to questions or otherwise as an outsourcing of their job. There aren’t enough GenZ in the workforce to back your claim in my opinion.
Token efficiency where instead of the AI burning money at 1:3 instead of 1:5 isn’t quite a winning argument.
There's no way managers using LLMs to answer emails are burning tokens at a comparable rate to someone trying to utilize inference in production systems is.
Maybe managers going back to coding.
There's the cost to do productive work, and the proportion of work which is actually productive.
GPT5.5 medium is ~20% the cost of Opus and 27% the cost of Sonnet on a task by task basis. That's a material difference.
Excessive token burning as a tactic to annoy your employer probably does the opposite - it probably makes your employer money.
The more tokens you burn, the more demand for hardware there will be. More demand for hardware means higher stock prices -> more money in your employer's pocket.
The only question is how long that can last. If taken to an extreme, the output of the AI will get worse over time, and if it gets bad enough, for long enough, people will use it less and less, and demand will slowly evaporate.
My point is NOT: I hope this all comes crashing down.
My point IS: tokenmaxing is bad. AND weaponizing it will not have the intended effect, but in fact, it will, in the short term, do the opposite.
> The more tokens you burn, the more demand for hardware there will be. More demand for hardware means higher stock prices -> more money in your employer's pocket.
If you work for NVIDIA, sure, but otherwise, this makes no sense to me.
I'm guessing you aren't invested in any of these stocks. I'm also guessing, to take the perspective of a tokenamxxing GenZ employee, that their bosses are.
Right, because every $100 spent on tokens results in more than $100 of profit from NVIDIA stock. That totally makes sense. But why so indirect, why not buy NVIDIA GPU and trash them? That should increase profit even more, since you're cutting out the middlemen.
All that aside, refusal to participate in bullshit doesn't imply some "plan" to bring the company down, it's really just refusing bullshit. It's not "5D bullshit", i.e. a move in the same bullshit game they are refusing
And if you offer someone 5 bucks to help burn down their own house, that of all their friends, so someone can make 5000 bucks, then they might just refuse, not because they have some plan, but precisely because refusal or letting themselves get hollowed out as well is the only option they're given. It's not like any of the hollowed out people listen to them in earnest.
Young people, which is crazy to me, often want some kind of stake and participation in the world rather than being disposable slaves, they want to do work that makes sense, that enables them and others to grow and prosper, in order to prepare them to become stewards of that world later on, which they want to make better and not just exploit, generally. Mostly, some of them of course are assholes or stupid, but in general, people come to this world way less fucked up than the world wants them to be.
Overheard recently: "Thanks to AI we're producing more code and more MRs, faster than ever, but the milestones aren't getting hit any sooner. Actually the opposite, if anything."
I wonder how widespread that phenomenon is. Perhaps it's no wonder the prominent actors are trying to rush to IPO...
Goodharts law. The metrics were always measuring the wrong thing, and now that we've finally optimized for the wrong thing successfully management will be forced to admit it and move on to another, slightly different, metric that doesn't actually equate to shareholder value.
It doesn't matter what the line actually measures, just that it goes up.
That's exactly what's happening. Many claim they are more productive with AI, but individual rise in productivity just doesn't translate to projects being completed any sooner.
And by "projects", I mean corporate ones with big teams involved. Hobby projects actually do get finished much faster.
In the corporate world, writing code was always a small chunk anyway. Iirc something like 30% of employee time. Getting 50% faster there still only gains back 10% of your time.
To further complicate matters, you had to spend on AI software and potentially additional on legal/risk/security/compliance to enable that.
The smaller the company, the bigger that % of coding is of total time (all the way down to hobby where the majority of time is spent coding)
My take from working at a Big Corp is that individuals using coding agents can increase velocity substantially and produce good quality work, assuming they are proficient with the tools. But it falls apart quickly when you have a team trying to work together.
imo we either need to centralize the agent and and submit plan, spec, reference doc MRs rather than submitting code changes. Or develop SCM systems/workflows that incorporate plan/spec/reference/prompt metadata with code so intent can be factored into merges.
We're planning for Q1 and don't have requirements yet for features, but are expected to have detailed estimates ready... This has happened before AI, this is happening after AI. The problem has rarely been "it takes too long to build ". The problem was "what do you want me to build? wait, no, not like that. now we must iterate"
I wish I could do the same thing. Coworkers would be allowed to ask me X number of questions per month, and once they hit that limit I get the rest of the month off.
lol. Solid idea. Going to add an email signature with "Emailing me is billed at the following rates: $20k/M token input, $100k/M token output"
Just a week ago, Anthropic barely breaking even was hailed as AI companies being close to profitability much earlier than forecast.
In fact it is all smoke and mirrors, pure mania from C-level executives out of their depth trying to one-up each other with company money, and they aren't even close.
I didn't look into the numbers myself, but Ed Zitron claimed in his podcast that the only reason Anthropic looks sort of profitable was due to some rather clever accounting (not illegal) and a spending that didn't happen, I think. Things haven't actually improved much, but if you zoom into a very selective part of the fiscal year it looks promissing.
https://www.iheart.com/podcast/1119-better-offline-150284547...
I really can't tell what is going on with AI these days. I hear AI labs claiming theyre profitable or close to it. I hear companies say they're dubious the juice is worth the squeeze. I've seen anecdotal claims of a measurable increase in productivity of 2x in PRs created and merged coming in at cost 20% of engineering employee budget. Others say they're still getting no value (which I doubt). Simon Willison's recent post went into debunking the AI sticker shock claims somewhat. Either way this seesawing between new golden era and the greatest VC money furnace is becoming exhausting.
I'd like to see real numbers at this point, and this article is just a few bullet points that link to other articles. Talk is far cheaper than tokens and I'd like to have a workflow that I can rely on being there in six months.
https://simonwillison.net/2026/May/27/product-market-fit/#th...
AI Has empowered people to build things much more quickly. Not slop if you are even a little conscientious about how you use it. What it does not do fix the human structural problems. If you are solving the wrong problems you aren’t doing anything useful. Just because you can now take an idea to near completion doesn’t mean it was worth doing, but now you spent tokens and a lot of your mental bandwidth to finish it. Or worse you let it become slop and it will fall apart if you even look at it funny.
Previously if I needed to automate something I thought really carefully about it. Now, I still think really carefully about it. I had fun AI coding some tools I always wanted but they were just pet projects for me. I had fun AI slop coding a couple of things, but it was not good software. But if you have a clear and valuable target? AI can absolutely get you there.
Multiply that across all your colleagues and a lot /seems/ to be happening, but what is actually moving the needle?
>Corporate leaders are starting to question whether soaring AI spending is delivering meaningful returns.
This isn't surprising. Ive recently run into quite a few rabbit holes where AI is bad enough that its much more efficient to do it myself. I wanted to refactor some code, gave it a design pattern to go towards, some specific classes and methods, etc. making it a well described problem. AI just couldn't do it satisfactorily. The code was ugly, overly verbose, and after multiple tries with multiple prompts saying to keep things simple. They still would introduce new classes, useless fields, etc.
"AI" is "bad" and "just couldn't do it"? Specifics w model and harness would lend more credence.
This was GPT 5.5 and codex. The specific model and harness isn't that important here. AI could do it. But the issue seems to be that there are some tasks where AI kind of falls over and provides poor results. It was easier, better, and faster for me to just do it myself. I have found a lot of cases where AI is great. If you have a UML diagram already, or translating code from language x to y, fixing unit test failures, generating boiler plate. But I can definitely see if people are using large amounts of AI for writing code, analyzing code, etc. that they are not actually seeing returns.
"An AI consultant tells Axios one of their clients recently spent half a billion dollars in a single month after failing to put usage limits on Claude licenses for employees." Like physically, how could this even happen?
I imagine some scenarios where you're throwing a big data set at an LLM could add up quickly.
"Check all of Confluence for outdated and conflicting info"
"Review all the legal contracts over the past 10 years"
"Evaluate the source code in all our dependencies for vulnerabilities"
All of those would be fairly straight forward for a single person to kick off without thinking about the data set size. Especially if they have Claude set to Opus 4.7 x high effort for everything.
Even someone saying something like "rewrite the 25 year old Java monolith in rust" as a PoC and leaving it running in the background for a week
Agents. An agent is a system for spinning up processes that use tokens, talking to other processes that use tokens.
I wonder how many megawatts that waste represented. Just one guy, worse than a small air force of private jets.
Now I understand the rumors of Anthropic starting to get profitable... by ralph looping fortune 500
This could happen with a large number of integrations or agent swarms working 24/7 with high throughput and long contexts.
Maybe the AI consultant gets a cut of that half billion?
Let's not forget about the environmental impacts. It's crazy that people are willfully burning so much energy for almost no return. And we thought bitcoin was bad... This is just completely irresponsible, if not sociopathic behavior.
Bitcoin doesn't have long conversations with you, including deeply technical ones, in a way that you thought only a human being could just 5 years ago.
They are not the same.
I agree fully with you on the potential for energy waste. We always do that, though, with nearly everything. How many of today's jet plane flights really needed to happen? The question is how much value people feel they're getting. People are having a whole lot more feelings about AI than they ever could about cryptocurrencies, and that train aint stoppin'.
I was specifically talking about needless token maxxing, though.
> How many of today's jet plane flights really needed to happen?
Many jet plane flights are indeed not necessary and there are people who deliberately avoid such unnecessary flights.
> The question is how much value people feel they're getting.
It is indeed a question we should keep asking and weight the answers against the energy footprint.
I am Jack's total lack of surprise :-/
How much longer before we get to the “I get cancer. I kill Jack” stage? Isn’t uncontrolled growth a sign of cancer? Or is it more virus like where it just continues to grow without ever reaching a balance with its surroundings?
>declare war on natural ecosystem
>win
This is because the current AI approach relies on AI to be a glorified search engine – know everything about everything requiring enormous, ever growing models, and demanding search-engine like near instant responses requiring bigger more complex chips and sprawling data centers to run them in. This leads to a loop demanding ever bigger models, updated at a more and more expensive cost, and chipsets that become much more expensive to deploy.
If you move those things to software and utilize tools that are cheap at scale (databases, web search etc.) the hardware arms race ends and the price becomes sustainable. With the right tools preparing dynamic context for a conversation, models are used for their reasoning and not for their knowledge. And waiting even a minute or two for a model to prepare a response, evaluate it, and iterate to improve quality makes a huge difference.
https://archive.is/crTG8
I’m also detecting a vibe shift in AI content
The mood has gone quickly from “this is cool” to “screw AI and any business that wants to use it”
This is particularly clear among the taste making class
Corporation always keep doing this, they hire lousy programmers then they try to make them produce good code by forcing them to use all kinds of hyped cutting edge technologies.
I'm sure there is an explanation for why they keep doing it.
> "Most people default to automating tasks they dislike rather than tasks most valuable to the company," Sophia Velastegui, CEO of Velastegui Ventures and former chief AI officer at Microsoft, told Axios. Instead, they should focus on using AI to drive revenue.
That's right! Work, slave, pain away every day! How dare you make your life a bit less miserable?!
"“It’s a real dollar investment,” says Tan, speaking for Claudeholics who go full blast. “You actually have to spend six to seven figures on tokens—I’m on a run rate to do seven figures this year.”"
https://www.wired.com/story/how-ai-agents-plunged-tech-world...
Has Wired gone evil now too? Just because some SV programmer circles who excel at marketing switched from Ayahuasca to a new drug does not mean everyone is going to use the new drug.
There are literally no good programmers mentioned in that article. They are all boiler room VC types. Shame on Levy, this article is planted or has has no clue.
The AI fever pitch has done a great job at exposing which companies were run with a degree of sanity, versus who bought blindly into the hype train narrative of worker replacement and went all-in.
Look, LLMs thrive when they’re given structured data that’s well annotated, clear direction, and treated as the probabilistic machines they are. Not one of those meshes with the AI narrative of “works on existing stuff, requires minimal guidance, and can behave deterministically.”
I said as much in 2024 when my employer at the time was grading folks on AI usage while my role was entirely deterministic in nature. It didn’t resonate with specific leadership then, it doesn’t seem to be doing so in the larger market now, and unfortunately not one of these dolts will suffer any consequences for their organizational myopia.
Maybe management is a more reasonable target for human equivalence ;)
I mean most managers in my incredibly subjective experience are indistinguishable from the probability slot machines that are LLMs, and those who weren’t were often managers I’d walk through fire for.
I’d love to see more research on the efficacy of LLMs as organizational middle management, but I fret that without sufficient anonymity protections for staff they’ll just do the same biased shit humans do.
https://www.merriam-webster.com/dictionary/sticker%20shock oh. i thought it was about people putting up stickers like https://mastodon.social/@spellingmistakescostlives@mastodon....
Just wait until companies are dependent on on it. When their employees can't think without it. When their AI generated codebase is such a mess they'd need a rewrite to understand it without AI. When they've got AI embedded in all their internal processes and tools. Then massive price hikes will come because they've been bent over a barrel and they'll have no alternative that isn't at least as painful in the short-term as letting the AI company fuck them. The long term won't matter then because any company capable of seeing past the short term wouldn't let themselves get into that position in the first place.
> Corporate leaders are starting to question whether soaring AI spending is delivering meaningful returns.
This is act one the AI bear market. Yes I know everyone screams “bubble”. Let me explain the scenario I have in mind.
1. AI booms because the technology seems to actually have promise of revolutionizing how work gets done. It can do your taxes! It can drive Excel! It can act as a CEO! It can code up full apps and SaaS products! It can replace this vendor or that! You know the drill.
2. Every company must in corporate AI or be seen as obsolete. Having a bad quarter? Announce that you are “seeking to explore opportunities to develop an AI integration plan framework” for your plumbing business. Massive AI compute buying happens. While two of the three major AI houses are not publicly traded proxies like Nvidia and RAM manufacturers are so the market rips higher and higher. Nvidia trades as if it is already 10+ years from now, every company out there has adopted AI perfectly, and it is delivering huge profits to them.
3. Reality checks start pouring in. Turns out that not only is AI expensive (a problem that presumably will be taken care of with time and development), but that the technology itself just isn’t suitable for everything. (IMHO it’s great at augmenting a power user but it is terrible at interacting directly with customers). We start seeing individual companies change tone on investment. They can’t stop it due to momentum but they are starting to shift the narrative to warn of what comes next. This is where we are.
4. Numbers come in. Earnings show what the actual ROI is. Some companies do benefit, but crucially we see examples of where investing in AI destroys value. I think this happens when replace crucial parts of their workforce with agents and find that they lost in-house expertise, when customers left due to worse products, or simply when AI was roughly as expensive as human labor without being significantly more productive.
5. The market stumbles. What do you mean AI won’t take over every corporate America?! Surely that can’t be right! Nvidia and other proxies flag.
6. In a late to the game rush Anthropic and OpenAI IPO fearing that the market has noticed that the emperor has no clothes. Their internal numbers turn out to be scary: very high revenue but no path to insane profitability. They quickly get included in QQQ and maybe even S&P500 but as their IPO price is the highest they trade they drag the broad market indexes down. This is leveraged by the Nvidia proxy status.
7. Infrastructure course correction. Hyperscalers who started huge datacenter buildouts cannot justify it. They pay contract penalties and get out of some of the projects, writing down losses. The market fully melts.
—-
I think there is a competing market downturn fueled by the affordability squeeze. Basically while AI spend is corporate driven, the biggest investors are consumer companies. Of the hyperscalers you can maybe argue that Microsoft is not a B2C fully but it is close. If consumers don’t have money to spend, hyperscalers take a hit, investment slows from there, and AI is hit directly by it.
I think either scenario is likely, it’s just which happens first. But right now the market is sprinting down a tight rope and trading like that tight rope has no end and that the sprinter never makes a mistake regardless of wind changes. Everything has to go right for a very long time to justify valuations. One stumble can stop it all.
> Instead, they should focus on using AI to drive revenue.
There is a complete disconnect between wages of employees and company's revenue => Why aren't employees working towards revenue? What a mystery. Children, let's help Elmo solve this mystery.
And then random mass layoffs to make numbers for shareholders look great in quarterly reports. Surely this motivates people work to their fullest potential and to care for company's revenue.
I found this claim interesting so I looked into it. Everything I can find shows that the intuition is accurate.
Companies with EOSP programs outperform those that do not in the market by about 17%.[0] Companies that perform layoffs, despite short-term stock boosts, underperform on a period of years showing a 14% decline in their Return on Assets (ROA) in the years following the layoffs.[1]
[0] https://www.nceo.org/employee-ownership-blog/new-study-shows... [1] https://www.researchgate.net/publication/277473996_Financial...
Even better there’s a complete disconnect between revenues and metrics we use to measure productivity. Corporate wants to believe there’s numbers you can use to measure knowledge workers like widget makers where there’s really not much that’s effective beyond revenue.
There's a very real possibility that we end up with a bunch of companies simultaneously realizing that they just don't have any good ideas to leverage AI on and rushing more code out the door doesn't lead to tangible value. Many execs are going to realize that they're no longer growth companies and that paying $500k a year for engineers no longer makes sense...
If that's the case, then I do expect the AI bubble is going to pop spectacularly next year as token budgets are going to collapse. The damage to the tech industry is going to be catastrophic. If you think the job market is bad now, wait until data center spending goes off a cliff.
These two paragraphs were paywalled for me.
https://archive.ph/crTG8
If I understood correctly, a few months ago Anthropic and OpenAI both started charging per-token billing at API pricing for Enterprise customers? i.e. representing roughly a 10x price increase? That's kinda nuts.
Similar discussion yesterday:
https://news.ycombinator.com/item?id=48296794
The classic blunder of believing that there is, in fact, a "free" lunch.
Well, if AI has a massive sticker shock attributed, so we should target the high value roles and should save money, right?
So im looking at CEO, CTO, CFO, and all the chief-something-officer. If LLMs are that totally amazing at thinking, then we should be targeting upper management, not the workers.
That would save a LOT of money for the shareholders! /snark
We all know why they wont.
So it goes. Wage theft dwarfs the amount lost to street-level theft, robbery, burglary, etc., combined. The economic stimulus from correcting even a portion of annual wage theft would represent complete coverage of those violent thefts - economically-speaking, there would be no reason for criminals to carry them out. Why rob a gas station to get your drug money? Everyone around you is making enough extra at work that bumming a dollar here and there covers it. That sort of thing.
But good forbid we actually correct a major social ill at the expense of the people who profit from it.
Exactly. And even worse is that crime itself is firmly rooted in politics.
Example: You work at walmart, as a cashier. You drop a $100 bill on the floor and pocket it from the till. You're caught, cops called, and you are arrested.
Example: You work as a manager at the same walmart. Store manager says labor costs are too damned high. So you go in on 10 employees, and edit their timecards to cut $100 from each of them, totaling $1000. IF you are caught, police will not respond. Instead, it is a "civil matter".
This is a bit dated chart, but its still very much correct. https://www.tcworkerscenter.org/2018/09/wage-theft-vs-other-... But notice the wage theft types are all "civil matters", and the non-eage theft are heavily criminally prosecuted. And who does those? Predominantly poorer people.
We also see this with dumping trash, being criminal and severe to the individual, but companies (read: Musk) can dump thousands of gallons of toxic sludge per day. https://www.yahoo.com/news/articles/inspection-texas-drainag...
We also see criminality differences between charges of freebase cocaine versus crack cocaine. Crack was what black people smoked, so sentencing was like 10x of freebase.
When you start looking at how laws are apllied, its almost always the same pattern: those at the top are a civil matter. Those who report to the top are a criminal matter. Those at the bottom are "charged with the fullest maximum punishment".
Right. Though I want to focus on a specific aspect of your hypothetical (which is essentially real):
If the manager hadn't stolen $100 from the cashier, there would have been a MUCH weaker incentive for the cashier to steal themselves.
This is the crux around which everything else turns: we are effectively post-scarcity, as far as production is concerned. As a society, we purposely create theft, and debt, and the associated desperation and crime, as a matter of policy. As a choice.
If you were eradicate wage theft, you would essentially eradicate the internal logic of street theft, in the vast majority of cases. We could just... not have theft. But by not prosecuting wage theft, we, as a society, have decided that we condone and even need theft.
Playing out in company after company right now:
CEOs: “Get me some of that GenAI”
CTO: “OK, we have all the GenAI”
CEOs: “Employees, it’s AI or bust”
Employees: Tokenmax
CFO: “Um, this is costing a ton and we’re not seeing savings or efficiency materialize.”
CEO: “Are we getting any value out of this?”
COO: “Not really, and frankly I’m getting annoyed at all the AI slop turning up all over.”
CEO: “OK, well, let’s do a big layoff and then I’ll just say it was because of AI. Hopefully folks won’t blame me for the mess and I’ll just talk about how amazing AI is.”
> Hopefully folks won’t blame me for the mess
Lesson #1 from business school : take all credits, put the blame on others, if there no easy scapegoat blame the "economical context"
Lesson #2 and beyond : just see lesson #1, it's enough. You've made it and it's ALL thanks to this amazing business school degree you got, now go profits with your new already wealthy peers.
I have a literal ledger tracking the number of times executives have used the phrase "evolving capital markets"
It's amazing to me how subtle the difference can be between a business leader that seems to know what they're doing and one that clearly does not. Same words, even, but from one mouth it's compressing a complex thought and from another it's word salad to give the appearance of complex thought.
Regurgitating word salad coming from Harvard Business Review or whatever publication is trendy at the moment.
Indeed the "worst" part is that the initial concept might very well make sense, even be grounded in actual research.
Masters of semblance.
These are golden times for AI consulting. We're getting into that business because it is up for grabs. Everybody as AI FOMO right now and begging people to take their money. Exactly what you are highlighting. And, yet, nobody has the beginning of a clue what to do or how to do stuff with AI. That's classic consulting territory.
My analysis of the whole space is that all the tech people are focused on completely the wrong end of the problem space (model quality, training, token cost, etc.). They are geeking out on that stuff. Worse, they are mostly not even that good at using their own tools. I regularly talk to so called "AI native" companies that are hiring enterprise sales people to do sales like it's 2006 instead of 2026. Their product is AI native, but their companies aren't.
The real issues are on the low level plumbing end. How do you connect all the data silos you have. Are they even the right silos still? And once you fix that: how do you get organized around tapping into those. What guardrails do you need? How do you deal with compliance issues? Etc. You don't need to do a lot to get value back. Most people haven't gone there yet.
Mostly people actually already have and pay for the tools that would enable a lot more if only they knew how: ChatGPT, Codex, Manus, Perlexity Computer, Claude CoWork, etc.
What you are outlining is funny but true. The issue isn't money or tools but figuring out what to do with the tools you already bought.
"Most people default to automating tasks they dislike rather than tasks most valuable to the company,"
Well, no shit, but also: suggests those tasks have questionable value? And also: this is why I learned to write code in the first place.