Companies are slamming the brakes on AI in a massive reversal that’s unlike anything I’ve seen in the last 25 years in tech.
6 months ago it was use AI all the time go! Now companies are putting use limitations in place, strict budget controls, and the wagons are circling around various “AI labs” teams that cost a ton and have shown little to no ROI.
It was all fun and games until the bill arrived. Now it seems there’s a mad rush for AI companies to IPO before the music truly stops.
Demand for a magic box that solves your problems at a low cost will always remain extraordinarily high. Supply is the hard part, because it will never catch up.
Some people believed LLMs were that magic box for a time, and that time is coming to an end if the parent poster is correct.
Just had to deal with this with a company that had outsourced its support to "AI". Probably saved them a ton of money not having to employ those annoying humans. Problem is that for this particular company once you get to the point where you have to contact support you're almost certainly in a situation that no stochastic parrot has any hope of comprehending, let alone solving. I spend about an hour going round in circles with the parrot until I finally figured out what to tell it to get it to give up and connect me to a human, who fixed the problem in about five minutes.
The scary thing here is that I know how the parrots work, what they can't do, and how to get around them. The typical person calling will assume they've been helped by the parrot, which is just going through the motions without comprehending anything or fixing anything.
Well, the history of cloud computing shows that infrastructure usually becomes cheaper over time. But it’s still unclear whether this rule applies to reasoning models.
Honestly that’s the trap that’s increasingly looking like it will blow up this whole thing. Nobody can point to any viable revenue pathway that justifies the amount of capital investment underway, all while folks are increasingly slamming the brakes on things.
Theres an extremely ugly financial picture developing that those with full blown AI psychosis appear unable, or simply are unwilling, to see.
Of course they can. They're going to sell ads and subscriptions. Both of which are going to make bank. That their service is wildly oversubscribed and hence expensive is not an indication that they're in economic trouble.
Ads are a zero sum game where there’s only so much ad money to go around. AI doesn’t grow the pot. Google isn’t going to lose the ad game, it would destroy them. Google got scooped early on with AI search but is roaring back now.
Also consumers won’t pay high amounts for subscriptions, that’s enterprise territory which doesn’t tolerate ads. And these are the folks now slamming the brakes on spending.
Net, “ad revenue” is not even close to a viable plan to save the present train from spectacularly flying off the tracks.
ChatGPT has like a billion weekly users that are giving them a massive amount of data. Everyone is going to want to advertise with them.
Enterprise isn't slamming the breaks on spending. At worst they've transitioned from spending like drunken sailors to spending like mildly inebriated sailors. Every single white collar worker is still going to have an AI subscription. And for people like programmers they'll still spend $1k on them.
Yeah, there's just this massive wave of AI delusion turning into disillusion. Writing code was never the slow part of enterprise development. We've made the slow part _somewhat_ faster, trading off quality in turn all while burning hundreds of thousands of dollars in tokens.
It's no surprise that when ROI remains elusive (it's hard to measure for any knowledge work) and costs are skyrocketing that the C-suite wants to slam the brakes.
Not my experience at all. One slow part was coding. AI takes care of that. But more importantly, the slow part was iterating through concepts, ideas, and prototypes. I thought people on this site embraced lean startups and agile development. AI really helps make that feedback loop 10X faster. I can do an experiment, show it to coworkers and get feedback in a morning, for something that would have taken me almost a week in the past. So now we can try a lot more options, whereas before, we kept getting hit by the sunk cost fallacy: I spent a week on this, I really don't want to start again from scratch with this other approach that may or may not be better.
The lean startup "feedback loop" was with customers (not coworkers). The idea was that you iterate on your viable product (not vibe prototype) with the market that derives value from it.
The slow part is finding those customers, syncing your deliveries with their processes, giving them time to meaningfully assess new workflows and features in the course of their business operations, collating the feedback you receive from all of them, and merging that feedback with your organization's long term growth objectives to drive new ideas into development. Well-developed organizations layer this inescapably slow flow across numerous parallel channels so engineering utilization can stay high since healthy engineering already cycled much faster than these market-engaged flows can.
Neither coding nor internal prototypes were the slow part. Market engagement and market-informed product planning were the slow part. And still are.
You may not realize it yet, and maybe you've just misrepresented it, but most of what you seem to be describing is usually considered wheel-spinning and navel-gazing. You may have made your internal process cycle faster, but you very likely just turned a wasteful busywork churn into a more efficiently wasteful busywork churn.
Neither coding nor internal prototypes were the slow part
That is not my experience mentoring 100+ startup founders. Building a prototype, the gateway to serious customer engagement, used to take months and many startups would die before finishing their first one.
Aren't those startups the ones wanting a google style infrastructure based on kubernetes with database sharding, an event-source architecture,... And when you told them a few VPS with postgres would have sufficed, they absolutely insisted that unless it's a next.js app backed by a serveless ecosystem and tens SaaS, they couldn't build their products?
Fair enough, experiences do differ. But how are you evaluating those POCs? Just based on 'visually what looks better', or architecturally etc?
In my experience, the slow parts are around making sure you're aligning on a long-term vision, understanding the domain and customer problem well enough, balancing the technical aspects/speed today with quality down the line, etc.
This probably does depend on what kind of tech problems you work on. If you're purely doing frontend development I'm sure you'll be faster. If you work on complexer systems with e.g robotics/hardware interaction, I can't see it being significantly faster. YMMV :)
There are multiple points of iteration. for me, it's user interface and core algorithms. Because the cost of creating an iteration was so high before, I would think about the problem for a long time and then implement the one that seems best maybe kind of?? I was always wondering that maybe I could have found a better solution. Now with AI, I can iterate through two or three solutions that I'm trying to decide between and see which one works best in a much shorter time frame.
You're not gaining any knowledge, insight, or experience from all of your iteration. You're churning for the sake of churn and pretending you're benefiting from it.
> I can do an experiment, show it to coworkers and get feedback in a morning, for something that would have taken me almost a week in the past.
That argument always rings hollow to me. What systems were you prototyping that took that long? I don't need to build a complete MVP to present a design. Or understand an API.
In the visual art industry, there are thumbnails and storyboards that are the first iteration of any project. They are quick to produce, and can serve as the basis for brainstorming. No one wants a finished picture, because it restrain your thinking. Too much details and you start bike-shedding.
Only when you've solved higher concerns and have a concrete direction that you start to invest physical efforts. But that does require someone to have the capacity to discern higher concerns from crude sketches. If you don't and rely on "I'll know it when I see it", then you sure need finished products to clarify your thinking.
Iterating and prototyping can certainly help there, but at the end of the day if you launch a non-working (or non-reliable) prototype, you’re going to just have angry customers, not happy ones.
And that rarely works out well long (or even medium) term.
And most of the value from iterating and prototyping is from learning, something the AI kinda screws with.
I've seen legitimately good outcomes with AI - a backlog has been cleared, features that were left on the cutting room floor have been pulled back in AND delivered all thanks to the use of AI coding tools. AI workflows have brought down processes from weeks of human processing to a couple of minutes with human oversight - and the revenue that it unlocks more than covers the AI bill. This is within a large corporate company - the "No such story exists for AI" feels overplayed. Sure, the wave of (quoting the article) "braindead executives, imbeciles and middle management hall monitors that don’t do any real work" might be bigger than with previous hype cycles because AI as a tool does enable pseudo-intellectualism, but the article overstates its case. I know, 1 counterpoint doesn't make a strong argument - but there's no reason the way we're applying this as a tool can't provide the same gains within other organisations - am I missing something/being delusional/huffing copium?
Yeah if your Csuite and managers are brain dead and pushing psudeo-intellectualism then how do the workers produce the same gains? My boss can’t even be bothered to project plan and half my company jumps at the idea of hiring a contractor for $15k-20k instead of understanding and implementing work themselves. Then cite efficiency as the reason, efficiency for what ROI?
I get where you're coming from, was shocked when I left a relatively well organised corporate and did work at a relatively older company with a ton of legacy systems - when I asked what the strategy was they explained the structure to me - at the year end results they highlighted that they hit targets of cost cutting and saw this as an achievement, the whole narrative was around how its a tough economic environment (the presentation was literally all about things happening in the world - nothing about things they did/projects they delivered/value they added...) - they also had more project managers than engineers and wondered why projects kept missing deadlines- they hoped AI would solve their problems - but you can't get ROI in a space like that where your engineers are using AI to patch the ship to keep it afloat while the project managers think they're in an airplane and are trying to get it off the ground...
> AI is more expensive today than it was three years ago, and it is not getting cheaper. Sam Altman’s comments about “intelligence too cheap to meter” were lies. NVIDIA’s Blackwell GPUs didn’t make it cheaper, and its Vera Rubin GPUs won’t either. Google’s TPUs won’t do it, Amazon’s Trainium or Inferentia chips won’t do it, Vera Rubin CPUs won’t do it, OpenAI’s chips won’t do it, and no, DeepSeek won’t do it either.
Has this man ever heard of Jevon’s paradox?
Also all of these claims are objectively wrong today because the goal posts for what AI have been moving this whole time. The models we have today do more, are faster, smaller, and cost less than what was available 3 years ago.
This is exactly what is happening right now. Models are becoming more efficient but at the same time users are starting to tackle tasks they previously didn’t even try to automate.
But Jevons paradox explains the increase in consumption, but it does not necessarily answer the question of business profitability
Maybe, maybe not. But his argument here is about cost, not profit. And on the cost front he is objectively wrong, and could easily see this if he cared to look. But he won’t since that would prevent him from making his bombastic content.
The author conveniently (or perhaps wasn't even aware) left out this quote from Uber's CFO
"What we have done is we have tempered the pace of hiring, and we -- and this is broadly across the company, but specifically from an engineering standpoint -- the hiring ramp we have for the remainder of the year is significantly lower than what we thought it would be when we came into this year."
Uber's response looks to be cutting the number of engineers that generate tokens, not to cut the AI that is generating them. These headlines about Uber are not the victory people are portraying it to be.
I've seen many cases where AI led to ROI with high margins (maybe not enough to justify the entire industry capex though), but they usually share similar features
- AI is a component of a larger product sold
- The product improves the metrics that customers care about, typically autonomously
- The customer is paying for the outcome, regardless of whether or not the product had AI in it
'Copilot' style AI features are much harder to measure ROI on, because they are typically further away from the base metrics that make it easy to measure ROI, and are typically used for specific tasks in a long web of other tasks within a professional job
Yes it does - the ROI is replacing the global labor market => the replaced workers stop earning income. They cut spending. The businesses they used to patronize see revenue decline => the company that fired its workers to save money discovers that its customers were, in aggregate, other companies’ workers. Revenue growth stalls => dead economy [1]
I'd probably characterise it as more as "AI doesn't have the massively transformational ROI that all the AI salespeople said it would and now I have to pay for my tokens and the humans I though I could replace at the same time". The idea AI would be running whole companies below some weird godlike CEO who won because they were clever just pushed an attractive narrative for the investor class.
I am very bullish on AI as a tool, but not as a way to completely restructure the economy overnight. Doing things is hard, and better tools don't make fundamental problems about change go away.
It seems to me that both sides are starting to drift into extremes. Some promised the replacement of half of office workers, while others are now saying that AI doesn’t create any value at all. The reality is somewhere in between
Also the article seems to be mixing two different things. The pricing as in
>allowed to burn thousands of dollars of tokens on a $39-a-month subscription
and whether the AI is worth it if you do pay what it costs.
I always thought the burn thousands of dollars of tokens without paying bit was unsustainable and harmful for the environment, electricity bills, investors and the like.
But I think it can do worthwhile stuff even if you pay. Like a small job I did was to get the emails of conference attendees off a website. It was tricky as they didn't want to be scrapped so I chatted to gemini and it helped figure out how to use tampermonkey and wrote a script. It was probably <$1 tokens and saved a couple of hours of mucking around. There must be a lot of things like that.
I am considering pinning whatever the earliest version in which this setting was introduced. I can't think of a single feature VSCode has implemented in the last three years that I couldn't go without. The binary for 121 is like 50% larger than 120.
Brilliant article. This is something I've been thinking about for a while. Up until around 2020 I used to work at a company that lived off of games and the economy was, you guessed it, micro-payments. I was the one in charge for developing the system that allowed the people in charge of monetization to configure the games based on your skill to squeeze the most out of you. Suffice to say, it worked great. Fundamentally the business model for all games was identical: cash for virtual currency. Here's the catch: you never knew if spending 50 bucks would make a big difference and you had no way to measure it. In a nutshell, it almost made a difference but just not enough so your brain would go "well what the hell, here's another 50" (classic sunk cost fallacy). And the business knew that and actively exploited it. All the AI slop that is happening now is the evolution of the same thing: exchange cash for virtual currency(tokens) in exchange for immeasurable results and the inevitable "just a few more tokens". Congratulations, you've been played.
6 months ago it was use AI all the time go! Now companies are putting use limitations in place, strict budget controls, and the wagons are circling around various “AI labs” teams that cost a ton and have shown little to no ROI.
It was all fun and games until the bill arrived. Now it seems there’s a mad rush for AI companies to IPO before the music truly stops.