AI is brains. Companies need arms and legs and relationships.

The reality is that employees are also mostly not entrepreneurs, same as consultants.

There are trillions of company ideas in those tokens, just like there are trillions of shots in the fans head watching the basketball players.

Takes years, patience, capital and the right mindset to get to the point to even take the shot.

The glossy magazines read by C-suite is the slick marketing part, not the actual product.

I think this comment was just a popularist repetition of discourse from social media. In other words it's a "tiktok ahh take".

GP is more cynical than I would have put it, but I think it's directionally correct; it's an easy way for the C-suite to demonstrate "due diligence" on any decision.

Accenture are mostly a bunch of low level talentless grunts. There are a lot of them. If you're hiring Accenture then you are in trouble in more ways than one.

McKinsey would like you to think otherwise of course. They paint a brand image of only doing high-end C-suite strategy work, but ultimately thats a vanishingly small part of the business. Their bread is buttered (just like Accenture) on labor heavy implementation gigs.

Fundamentally, the business of billing out human hours is the same, as are the the incentives. Doesn't matter if its a strategy consulting firm in NYC or an advertising agency in Nepal.

They will all inflate scope, inflate titles/rates, hire mostly cheap new grads, and slowroll-and-expand any engagement.

And they will all take ANY work you give them, regardless of in-house expertise (which they rarely have, given the goal is to pay employees as little as possible to capture margin).

The advertising agency in Nepal will claim to offer strategy consulting to your C-suite, just as McKinsey NYC will happily churn out a powerpoint deck about advertising in Nepal if you're willing to pay them.

Hence why professional services firms all trade at low multiples. It's an undifferentiated commodity business of re-selling human labor in project-based chunks.

In fact, the bear case on Palantir is always that it's actually a consulting firm masquerading as a Saas firm.

Plumbing is fractionally less regulated where I am and I am allowed to change my own taps washers, but a whole tap assembly falls under similar issues as a power switch.

"So the third party consultant warned against it, you did it anyway, and it caused these problems?"

When you say [insert firm] is [good/bad] at [thing], what you're actually saying is:

"The team of kids at the firm's Minneapolis office did a good job on this specific project."

There is no special sauce or "process" that could consistently deliver that outcome given consulting firms are just hourly billing body shops with high turnover that rarely hire true subject matter experts (they don't pay well enough). They have no equity in the advice they hand out, they don't suffer consequences if they're wrong absent doing something illegal, so nothing is ever learned other than how to extract more projects from the next lazy corporate bureaucracy.

It's wholly dependent on the kids who were staffed on the project and how insecure/eager they were for approval at the time.

Average tenure at all professional services firms (doesn't matter if its consulting, advertising or accounting) isn't very long, and there's almost zero knowledge sharing between individual offices of the same firm since they are competing for billable hours. So generally it's a crapshoot every time.

What you can consistently be sure of when hiring McKinsey, is that you are offloading responsibility for decision-making to a 3rd party fictional mirage of 'competence' that everyone pretends is real, so you can claim its not your fault if something goes wrong or is unpopular...and keep your cushy big company executive job.

We used to have a very experienced systems engineer that we'd contract whenever a chief engineer had an idea that gained traction with directors. The systems engineer would help layout the structure and timeline of the project, basically identifying how much money and the types of people we'd need. He'd stick around long enough to get the project to the final design review and then go do something else when we started building stuff. By that point, the PM is deeply involved and the team is cohesive and productive. Maybe it's just a coincidence but every big project we've tried to stand up without his help would struggle tremendously in the early stages.

That said, I would be highly skeptical of any 'very experienced systems engineer' who took a job at a glorified post-collegiate body shop like McKinsey. At least in the US, given all the opportunities available, it means something has gone horribly wrong (EU/ex-US might be different given it's a dead zone for tech).

Is it smart for individuals to have conversations with others about their problems? This is that for CEOs and senior executives.

Bigtable these days handles seven-billion requests per second over several exabytes of data. All developed in house.

The others are similarly impressive.

https://en.wikipedia.org/wiki/MapReduce https://en.wikipedia.org/wiki/Bigtable https://en.wikipedia.org/wiki/Spanner_(database)

Total or Shell use a lot of state of the art technology to mine, they are still oil companies.

If an oil company creates new technology to extract oil, then they are (obviously) a technology company. If an oil company outsources all technology development, they are not.

That’s especially true if it’s an artificial gold rush that you created yourself.

OpenAi is also spinning up and acquiring companies. DeployCo, Orchestra Bio, io Products Inc.

These elite consulting firms are all about being the fall guys for things that would cause employees to start talking about unionization. They also exist as a jobs program for the kids of very wealthy people.

The Firm is confederated.

Sorry, this is not even close to being an answer let alone an obvious answer. The main reason is because 'reality has a surprising amount of detail' (read the excellent blog post if you have). There is a huge difference between doing some analysis, framework, and outlining recommendations on a slide vs. generating business value in the real world. The real world is messy, a complex and dynamic adaptive system, with tons of details where small decisions add up.

Put a feature team in a strong company / startup and put them up against a top consultant on a real world scenario to ship something and generate real world value/uusage and you will get the answer why.

It's a serious bit of hypocrisy posed by OpenAI here, and it's telling that the AI is much more limited that the narrative of 'AGI' would have us believe.

OAI and Anthropic have massive budgets they should be able to expand, were AGI actually AGI.

That's a large chunk of the software development industry, but practically nothing else.

So they promise robots, but have remarkable difficulty with even the most limited domain autonomous robots: Amazon claims to have deployed over a million robots, but is still continuously hiring in their warehouses with requirements like these:

Be 18 years or older

Understand and follow instructions in the required language

Lift up to 49 pounds

Push utility carts up to 60 pounds

Stand, walk, push, pull, squat, bend, and reach during shifts

Use carts, dollies, hand trucks, and other gear to move items around

Go up and down stairs (where applicable)

Maybe Amazon shouldn't have built anti-Dalek defenses into their warehouses.

https://www.gatech.edu/news/2026/04/01/why-iran-targeted-ama...

https://arstechnica.com/gadgets/2026/09/iran-strikes-on-amaz...

A lack of anti drone defenses turned out to be an actual problem - also interesting use case for your idea of durability to be judged against.

Give it 5 years and see how strong those anti-Dalek defenses are.

FWIW no I don’t believe models can replace intellectual work because intelligence is not purpose, and no other entity can tell us what we want. While AI has made building cheap and cutting through the noise even harder, it hasn’t made deciding what to be build any easier, and personally I felt that was already the bottleneck a decade ago (think bullshit jobs, App Store saturation, etc).

Of course that’s just one perspective and I could be totally wrong. But given the situation it makes perfect sense for the big labs to stay focused on building the most valuable shovels in the world and leave the business risk and finding the money for the cash furnaces to others.

That's the marketing department's job. At some point, there will be an attempt at modeling that, too.

We live in a time and place where the only way you can truly profit is to cut costs. Selling more of the thing that you make or do to more people is harder because of population trends. Everything that can be modeled in AI and done by it, will be.

Thinking our governments will solve the issues the AI vendors created ain't going to work.

The former Starbucks CEO is a classic example. His infamous interview with Cramer on CNBC is the textbook of what it looks like when some McKinsey consultant gets actually put in charge. It was a total shitshow.

Likely succeeded at pontificating to CEOs via PPT, but actually being CEO? Not so much.

It opens with the statement “As it nears 100, is the world’s most illustrious consultancy past its prime?” and cited poor performance associated with its advisors. So your “citation” isn’t really helping the point you’re trying to make.

Do you have a link to the specific interview you mention? Google seemingly doesn't know of an infamous one.

Absolute car crash interview that eventually led the board firing him.

Is they though?

Do they? Haven't Google and others been proving that wrong for almost 30 years?

Business is not just person to machine or machine to machine, a lot of it, even those of you that are in pure tech jobs has someone in the field selling to clients and keeping clients engaged.

Those are very much not the same.

This is silly, these companies are owned and run by some of the most connected and powerful people on the planet, who also own and run dozens of other companies that produce things.

> The reality is that employees are also mostly not entrepreneurs, same as consultants.

I don't even know what this means. They choose to hire who they choose to hire, and they can continue to hire as long as they wish. It's like saying that the reason I'm not a marathon runner is because I only own Oxfords and flip-flops, not running shoes.

I have absolutely no idea how this could be the "sort of obvious answer." I'm surprised so many people thought it was worth a response. There's no answer here. Some of these sentences don't even make sense.

2 things: 1. Yes they are. Yet here we are not using WorldCoin despite its obvious utility. But backdooring every major software library in the world is pretty useful for governments. Too big to fail. But not an economic weapon, more like a cyber one.

2. Employees and consultants don't have what it takes to start a business, that's why they become employees.

Just take a second. I promise it makes sense. Many see the simple point I'm communicating.

Employees have everything else they need -- intelligence and the ability to put in the hard work. The only reason they don't have the capital and connections is because the existing "entrepreneurs" gatekeep it to minimize their competition.

The landed elite will continue to defend the only moat they have -- capital and connections. They know they're not the smartest or hardest working people in the room, and if they start to recognize their employees as equals they'll quickly lose their position within their company and within society.

For AI companies to cannibalize the rest of the market, wealthy investors/executives first need to admit that they don't have much to offer society. It'll take a long time for them to allow this to happen, lest their egos (and portfolios) take a hit.

Also the reputation is not minor, the human can take out loans, or take a withdrawal out from their social capital by telling people that they are AI enabled and selling AI to other (presumably) non-AI people.

It's not all just GPUs and Tokens, it can't be, there needs to be some mixing with other resources in order to bake a value cake. While pure token cakes might be very yummy (for some consumers who like pure slop), by mixing GPUs with human brains, a wider space of yummy hybrid cakes can be explored, which even more consumers would enjoy (those who like partial slop).

Even drug manufacturers need retailers who use their personal connections to gain trust, and who put their criminal record on the line.

I don't really know if you could do multimillion dollar deals on a web form.

>I think this is basically what it will be for most knowledge work, knowledge workers are so grossly overpaid compared to the energy they consume, and AI will rectify that. This explains why the Chinese are giving the (much more moderate resources to train) models away for free. They love to see deflationary economics in the US.

https://geohot.github.io/blog/jekyll/update/2026/06/11/ai-wi...

"knowledge workers are so grossly overpaid compared to the energy they consume, and AI will rectify that"

Then what's left to work on that pay well? machines already do things that depend on mechanical work faster...

Is this the end-all argument that AI is really a mean to pay the least possible to people?

a.k.a. the ultimate transfer of wealth and power to uber rich people?

I think it's less that they want to encourage deflationary economics - because I think this has a tremendous chance of backfiring given that is not how massive productivity/efficiency gains have turned out in the past.

I think they genuinely want to prevent the rise of yet ANOTHER Western bigtech monolith in Anthropic and/or OpenAI. They want to make sure they have no moat. That keeps the value out of the nascent bigtech and instead in people who can run the models.

Add to that the fact that the AI itself is going to be more or less a commodity as well and you wonder where all the profits are meant to come from.

LLM makes may make more money before their tech really works than after.

Yet picture is very sad , industries affected to the point that majority of the companies asking about their existence:

1) software dev tools all IDE , developer analytics tools eg jellyfish etc . Dead . On the same note : warp.dev that raised $80m a year ago as an ai terminal… can barely explain why do they exist now.

Visual workflow editors . Older ones and even n8n. There were literally 1000 companies that were doing small deployments. Why do you need it now ? A year ago half of the Internet was booming about these tools .

Any data visualization tools . Gone .

Data integration tools all kind of crm to crm to erp to something else . Solved problem

And you can go on and on . I would probably not exaggerate if I’d say that millions of businesses are heavily affected by, from tiny one to adobe .

There's no reason for OpenAI/Google/Anthropic/etc to start other software businesses because "LLMs" will be the only software that anybody will ever pay for.

Why would you have a different "app" to do something when you can just do it all in Claude or Gemini or ChatGPT? We'll be there soon enough.

Twitter consensus now is more like LLMs will be entirely commoditized, apps & features will be cheap, and only systems of record will have lasting value.

Non-deterministic behavior from your chatbot, like how the visual design artist's client keeps sending back and forth emails asking for it to "pop" more

why are you comparing to your buddy? we're talking about software products?

That's where we're heading.

Perhaps even by other people, and over time, continuously improving as we discover new use cases or that things are more complicated than they seemed on the first pass. Like... regular software, and not "do[ing] it all in Claude or Gemini or ChatGPT"

“Claude deploy this to dev”

vs

“Claude build me a tool that allows me to deploy to dev with a single click / command from CLI”

Orders of magnitude less expensive for the latter than the former. The former usually costs a few cents and the latter maybe a one time cost of 40 cents and then free forever afterwards.

You're thinking like a developer that needs to build and deploy software for other people to use, rather than as a user that wants to do something.

If you just need a narrow subset of the functionality for now, might be cheaper in tokens to just vibe it than the attentional cost of searching for just the right software...let alone paying for it.

[1]:https://chatjimmy.ai/

  - stuff that’s easily replicable that I can tailor to my use-case (e.g. I have an alternative to git add -p that I generated that I find much more ergonomic)
  - stuff that’s a giant boondoggle to replicate because there are so many edge cases and considerations that it takes years of iterations to arrive at something stable

Am I really going to create a git alternative? Who in the right mind would take the time to even learn it?

And also, no, Claude isn’t going to replace git. Deterministic compute still has a place in software, nondeterministic compute isn’t going to take over just because the UI is more pleasant for nontechnical folks.

Well, If you're an iTerm2 user still after about the 1st hour of setting up a new macbook then you probably don't have need for the features that customizing a terminal to your own needs brings.

Claude will totally be used to build a tool to replace git for specific groups of users. There are alternatives popping up left and right, there was one that was announced by Anthropic itself last week.

Clause doesn't need to be deterministic when you can use it to generate software that is deterministic in its output.

I mean this is my point in a nutshell. Why do I even open a chat window if I already have a tool that does what I want it to do.

Ultimately the value will be in the data.

We're in a window here where the data is silo'd and we can use agents to vibe code / glue it all together and create traditional apps that people will still use, but I think that window is narrow.

Eventually today's LLM providers will incorporate direct access to more data sources, and the data won't be silo'd for them - it'll all be available. And instead of third parties vibe-coding apps, the LLM companys' web UIs will become the apps that link everything together directly.

And the compute.

Anthropic, Google, OpenAI, Meta, xAI(whatever is it called these days) are sitting on mountains of compute.

Yes, local and open-weight models will continue improving, and they'll be good enough for an increasing number of tasks. But there's a difference between making yesterday's capabilities cheap and pushing the frontier forward.

The most capable models will likely continue to demand enormous compute resources, both for training and inference.

And the pricing is already there - we're charged for token usage, not model access. Draft an email, summarize a paper - basically free. But the "big stuff" will require Frontier-models with Frontier-level of compute.

OpenAI spent something like tens of millions on "solving" a Millenium Prize math model. And there will always be another problem worth throwing massive amounts of compute at, whether it's mathematics, scientific discovery, engineering, etc.

I keep wondering if throwing tens of millions at human mathematicians would have gotten similar results.

Let me go on a tangent to make a point. For some countries if you want a US Visa Consulate appt today they only open 3 months out and are booked constantly. They block you for 10mins on the 5th refresh.

It’s well known on crrtain communities that the only way to get one is with an ai agent that can patiently work around this and jump on the bookings with perfect timing or refresh within the bounds of the detection to nab a cancellation opening.

So we’re starting to head into a world where every single person needs an ai subscription just to deal with bureaucracy.

It’s a fucking fantastic business and i don’t understand how you could suggest it’s not good business.

If everyone uses an AI agent as you suggest, the appointment schedule will be bottlenecked just the same as if everyone only used a web browser. Using AI won’t achieve the goal of everyone getting an appointment. So AI won’t actually be useful here if everyone uses it.

At least from the engineering perspective, the app has a massive advantage: determinism. If I write something in Python or Java or whatever, it has set inputs and outputs and those can be proven to accomplish a computational task because they are working in a logical system.

If I ask an LLM to do that same task, it has inputs and outputs, but the logic isn't anywhere nearly as defined. There will be a chance, no matter how small, that will not act in a way that makes any sort of logical sense given the constraints of the task it was meant to accomplish.

When you combine that with the fact that deterministic applications are far cheaper to run, at least for the foreseeable future, the app wins.

There are usability and performance problems right now which is why we have this interregnum. But as these systems evolve, the tools can offload "expensive" operations to traditional compute by writing the glue code on demand, and persist the code for future use (so regenerating it won't be required again later) when the user confirms it's doing what it's supposed to do. You can A/B test it in real time and ask the user to pick the "better" result.

Same is true for e.g. GUI elements, widgets, etc. Write it the first time, have your user verify it, persist it, retrieve it.

And maybe you have a couple of people in your org who work through the process of generating a bespoke app that processes data with a GUI optimized for whatever they're doing, but it's all mediated directly through your vendor. The users run through the process all in the web UI of whatever LLM vendor the org's data is silo'd with, the output is a mix of traditional computation and LLM contexts that anybody can maintain via asking the tool to change itself.

I might have the general desire to play a video game with my friend where we shoot WW2 like guns together. I'm willing to bet that a video game company with artists, game designers, historians, etc. is better able to articulate that into existence than I can. Especially given I might not be willing to spend the time to do that...

this is a point that i think is missed on many people (myself included earlier in the year). Sure, the LLM is non-deterministic but it can write code that is not and test cases to verify it. Then, once verified correct, it can use what it wrote to perform the task in a deterministic way. Even though the LLM is still non-deterministic all it has to get right is the arguments to the software which is much easier than the full logic of the task. That's very powerful.

I've seen this first hand when asking AI to do non-trivial analysis over a couple of spreadsheets and then also a database. It writes python to do this, tests it with sample data, and then calls the python it creates to do the processing on the actual data.

How is this different from what we’re all doing in software development right now? It’s just we review and make the code more robust, and build it into more complex systems. If you have repeated workflows, it would be idiotic to just throw away the code that was generated by the LLM instead of properly building a system. Sure, sometimes you have a one-time use-case. Most non-technical people do. But most technical people are in the business of building systems these days.

It will all be one integrated UX soon, as the LLM UX best practices are discovered.

Have the LLMs generate even more deterministic software to solve well defined needs and keep the LLMs for solving higher level problems using the deterministic software's outputs as inputs.

We have had AGI for years now.

If I need to process a million items it still wouldn't make sense to feed them directly to the model and burn a crazy amount of tokens

Of course the model can produce code to process them on the fly but if its a repeated task why would I want that to be re-created every time, especially if I've already tested/validated it

That overhyped thing that never did anything well and everybody that used them regretted it later? Yeah, hyped things tend to die once the money used to hype them ends.

> Any data visualization tools

Microsoft killed them with anticompetitive practices. It helps that every single one of them tried very hard to kill themselves by being as hostile as possible.

> Data integration tools all kind of crm to crm to erp to something else

Yep, everybody writes those in Python nowadays. Apache is a great market killer, the best kind of market killer.

I'm sure there's some impressive stuff in there that the warp team built but also the second my own _terminal_ tried to get me to subscribe to a monthly plan I uninstalled that shit so fast.

Of course you shouldn't just rely on AI that's consumed every MBA book to give you all of your business advice. Sometimes the textbook response is not always the best response.

Now with LLMs that's all solved.

IDEs will be moot when I die. Even with tests and code gen, I still debug and verify code before shipping to QA.

This is a wild take.

I read the merge request after validation, testing, then only for verification before shipping.

The perpetual state of being in the inflection point upon which entire industries are just... killed.

https://ghuntley.com/kodak/ — there's no inside information here, but the article is spot on.

The article doesn't mention this at all. It just blindly includes that graph and than muses about the death of Jetbrains for a few paragraphs.

I actually disagree here, what matters most is procurement & distribution. You can sell absolute dog water software (see: Teams[1]) and capture a market with the right procurement and distributions pipelines.

The sad reality IMO is that the quality of software matters much less than we (the engineers) would hope it does.

[1] https://web.archive.org/web/20260322032609/https://www.theve...

The issue with that belief is that it's basically impossible to falsify.

The right framing is ecosystems.

You don't think AI is capable of putting out CRUD apps into the ether that work reliably? Even my local GLM5.3-Flash is capable of doing that. These products make up a non negligible amount of the industry. Additionally reducing headcount is about to hit in the next year or so. Where are juniors gonna find jobs? Etc etc...

Sure thing, I can see any sort of serious discussion is off the table.

Spec -> Prompt -> Validate -> Test -> Merge Review -> Production -> Agents monitor production -> Alerts -> Roll back

This is true for a many of us. All my team's code is now AI written.

im already practiced at reviewing code through code review tools, so id rather do it that way than in vscode

The translation market is also wrecked right now. Nobody needs a translator to Polish, the LLM is really good at that.

And that's not counting AI companies "cannibalizing" their own products, like Sora video generation

2. Selling tokens via long running agents is the only thing that's provably working with respect to revenue growth.

3. They need revenue yesterday. Increases in borrowing rates means access to capital isn't what it was in the 10s. Their runway is much shorter than you would imagine.

4. Some of the decision makers at these companies believe AI will exhibit Jevons Paradox and that any decrease in price will result in induced demand.

5. The frontier models are still subject to the limitations of AI. They're not good at everything. Better to have someone else learn that on their dime rather than yours.

Openai haven't yet amortised their current investments into the space and nor have they stopped making new ones.

I'm sure they'd be totally capable of bringing 1k new biz online but that's even more debt, and even more uncertainty.

Although the point would be equally valid if they were OpenAI consultants. (Imagine paying twice for the help: once for the consultant and once for the tokens.)

If they don't make it, there will be plenty of time to mine through their dataset for business ideas to sell to other people as they liquidate. Think of the Spirit Airlines dataset that was just sold.

That all said, A person trusting the big AI companies with their golden goose idea seed leaves a lot of concerning openings down the road if that idea gets "leaked" out to others. Similar concern of mine with if the LLM outputs that made your idea happen to be reused by the big company if someone else asks similar enough questions or otherwise coaxes the AI to reveal what it did for you.

All this AI generation is gonna be a lead to a wild time for IP law.

This is a huge problem, why is it like this

Investment isn't debt.

Presumably there are penalties if OpenAI were to cancel that order (I am confident that Oracle's army of lawyers made sure to include that). That's the unmovable rock on OpenAI's books, not the data center.

Not a useful term, here.

AIUI, the building is funded by making the people who order it part owners in the new data center company. Not sure and I don’t care enough to find out; clarification welcome.

One of the biggest AI investors.

Does SoftBank have some absolute blunders on large bets? Totally. Do they make some super weird billion dollar bets? Absolutely. Can you scrutinize them for making some terribly and laughably large bets go to complete zero returns? Sure. But this would be like saying Michael Jordan had a "bad track record" for scoring points because he had games playing for the Wizards where he had like 5 or 10 pts in a game.

When you can throw billions around for investments then on balance you will make a lot of money because you can take a lot of risk. But it does not say anything about any particular investment and if you look at the ratio of good bets to bad bets then SoftBank is terrible. That does not mean they won't make money or that they won't be able to keep doing this. Capital - especially at this level - begets more capital.

How does this explain a $20M bet on Alibaba that netted $60B?

> if you look at the ratio of good bets to bad bets then SoftBank is terrible.

So does a16z, Sequioa, etc...

This is venture investing 101 - they have power law returns. Don't let your survivorship bias cloud your vision.

It's a 26-year-old investment. The advisors who suggested it are long gone from SoftBank, and the stake in Alibaba was sold to provide liquidity for bagholder investments like WeWork.

SoftBank was lucky, and now their luck is gone. Son had an investment in Nvidia that would have been worth $100+ billion, but sold it in 2019 for only $3.6 billion. Uber is struggling, Arm is a laughingstock of a business, their stake in OpenAI is comically overleveraged, and Softbank Vision Fund is one of the most heavily ridiculed and mocked international VC funds anywhere in the world.

We have to move past a decades-old bet, SoftBank's current state is not reminiscent of their prior investments.

AnthropenAI are the ones driving a lot of that infrastructure spending when they’re such big consumers of compute but they’re not the ones feeling the brunt of the infra spending. Although those costs will bubble up to them indirectly eventually.

So in effect the debt is backed by OpenAI/Ant's future revenue stream.

In part it is like this because they've already borrowed so much.

Investment isn't debt.

No they don't.

A company has legal obligation to pay loans at the agreed rates.

There is legal obligation that defines rates of return beyond "acting in shareholder interest" (and OpenAI is a public benefit corporation so even this is reduced).

Because it's an investment it is risk capital. They won't be happy if they don't get a return but their return is part ownership in the company, not a legal contract giving an income stream.

That's what is driving up the cost of debt and teetering some of these companies into junk status.

No.

> If VC provide you funds, there is an expectation of returns, otherwise the board would vote you out.

A person might get replaced as CEO but the company has no financial obligation to the investors.

A loan means the company does.

Legally they are dramatically different.

There just isn’t a sane way to look at this as “investment”

Causality is uncertain, but the "Jevons paradox" interpretation is certainly not contradicted by the available data. Anecdotes of per-employee AI spend rising, and multi-agent freerunning AI tasks becoming more common are also plentiful.

Personally, I think that the "Jevons paradox" interpretation is basically correct. Humanity doesn't have a limited demand on intelligence - but humans are expensive, so it's supply-constrained. The cheaper intelligence gets, the more of it ends up used. As AI gets cheaper and more capable of substituting for human intelligence, usage will rise.

Because building the product is the easy part. Most products and businesses fail because they build the wrong thing, have the wrong requirements, have poor market fit, etc.

LLMs can write code all day, they don't have the agency to know what to build.

Hence this is also why although AI is writing the code now, it doesn't replace the companies or build entire large scale products without human oversight.

Agent swarms, you say? Perhaps. But I don't think we're there quite yet. At least, we're not seeing it manifest.

https://apnews.com/article/ai-artificial-intelligence-sweden...

> San Francisco-based startup Andon Labs has put an artificial intelligence agent nicknamed “Mona” in charge at the eponymous Andon Café in the Swedish capital. While human baristas still brew the coffee and serve the orders, the AI agent — powered by Google’s Gemini — oversees almost every other aspect of the business, from hiring staff to managing inventory.

https://futurism.com/artificial-intelligence/google-gemini-r...

> Some of its questionable business decisions include ordering thousands of rubber gloves, despite the café only having a handful of employees. The AI’s handlers, nonetheless, are holding out hope that this is just a blip from expensive setup costs. How well it performs will raise grander questions of the tech’s impact on the workforce.

> “AI will be a big part of society in the future, and therefore we want to make this experiment [to] see what ethical questions arise when we have AI that employs other people and runs a business,” Hanna Petersson, a member of Andon Labs’ technical staff, told the AP.

> To launch the experiment, Mona was given a simple set of instructions. It should run a profitable café, be friendly and easygoing, and try figure out operational details by itself, Petersson said.

---

And that last part is the "why doesn't a LLM company open up a company to sell its wares". There is knowledge from life experience that a LLM doesn't have.

For a product, you need to recognize a need in the economy... something that irks you that you can make better. Go through https://www.npr.org/series/126166186/from-scratch or https://www.fromscratchradio.org/show/ (or any other podcast service) and consider how often those companies were founded on something that was recognized as a need or something missing that comes from the human experience ( https://www.fromscratchradio.org/show/mick-mountz for Kiva systems ).

---

The company represents risk. The coffee shop needed money to start... and it failed. https://kotaku.com/google-ai-put-in-charge-of-swedish-cafe-b...

Creating companies making applications that people don't need is wasted money.

Proof: - Load up all your chats, across several different topics and see how even the best models will try to connect dots with lines that don't make sense.

As a side effect of the above, LLMs have trouble coordinating with other LLMs.

Humans are able to compress and manage context without much effort and each human brings their own unique set of experience and perspectives.

That's why humans scale in ways that LLMs cannot yet match.

Maybe by next year they'll crack this too but that's where we are today.

Didn’t DeepSeek just disabuse us of that notion?

> …with overall revenue of US$38 per GPU-hour (at valley pricing)

https://github.com/deepseek-ai/DeepInfer/blob/main/doc/infer...

At the top end B300s are $9/hr from RunPod. Now think about large volume discounts etc

Does it mean 'exactly human level capabilities in all aspects that a human can perform'? Because I agree with you on that point - no one really wants it, either. We already have several billion computers in the world requiring a low power requirement, and fairly easily reproducible, that can do that.

Does it mean 'smarter than a human on most tasks'? I'm pretty sure Opus 5.5 has crossed that line fairly definitively.

Does it mean 'smarter than a human on all tasks'? I think it's doubtful we'll ever get there with LLMs which are trained on human-derived data. I also generally think this is "ASI".

https://www.youtube.com/watch?v=ZmInkxbvlCs&t=10s

And if the AI isn't general enough to handle every part of a business well then you will not convince people that its AGI.

This hasn't changed from 10 years ago, this level was always what people imagined about the future of AI.

> a Competent AGI must have performance at least at the 50th percentile for skilled adult humans on most cognitive tasks

and "emerging AGI" as

> "equal to or somewhat better than an unskilled human"

The moving the goalposts here is people saying it has to outperform humans on literally every task to be AGI. We have artificial general intelligence by any reasonable definition; people here are complaining we don't have super-intelligence.

[0]: https://deepmind.google/research/publications/66938/

Down the chain you see what you need to do because you work directly with it everyday.

Neither would be easy for either to do

But look at the ArtCraft project. At least one of labs has an internal benchmark to clone 20+ popular products much like ArtCraft but with much higher budget and several configurations of the workflow. ArtCraft spent $20K initially, not sure how much by now, but they are moving fast and the apps work. They day is coming when they can flip and just clone everything including the ops.

I find Claude Design to be really interesting. It made a huge splash, and now...nothing. I believe it hasn't been updated in a few months. I read a comment here that said that it was like the person who spun it up had moved off the project or abandoned it. Keeping in line with the thread topic, perhaps Anthropic is just biding its time until making humongous updates that will vastly increase its capabilities?

The downfall of Apple may be closer than people think.

They don't have to do it openly.

Nobody "shared".

Citation needed.

Why doesn't NVIDIA just make the best frontier model?

Why doesn't TSMC just make the best chip design?

Why doesn't ASML just make a fab?

The real answer here is that OpenAI and Anthropic are in the middle of an intense race, and the payoff of having the best model (or one of the two best) is so large that it makes no sense to focus downstream.

The barrier is often that the investors aren't fond of the heavy investment and risks involved. Using AI to generate some app would presumably not have the same level of concern.

To me, this says something about the level of expertise they offer.

I’m a senior developer, I use Claude every day and am amazed at the quality.

But I also see the shit that comes out of Claude when some colleagues, who were poor engineers to begin with, drive the AI.

It’s not just about the model IME. How it is used, really makes a difference.

I think this is a big part of the answer: the human driver brings judgement and taste (IME, YMMV)

Yes. in a gold rush, the best strategy is to sell shovels. they sold us an idea of whats possible, while they earn NOW.

come to think of it, whether you succeed in building the "next big thing" with thier tokens or not, they get paid.

But where the business model lands we simply have no idea of currently because the future is in the future.

Except they're not. Those shovel sellers weren't losing $2 for every $1 shovel they sold.

Secondly even they are constrained by resources. They have a limited amount of people/GPUs/capital. Given the constraints its best to focus your attention on the highest ROI given your team and organization.

https://www.smbc-comics.com/comic/2010-01-29

https://www.yahoo.com/entertainment/tv/articles/televised-ne...

I think OpenAI actually will ultimately start a fleet of consultancy/SaaS spinoff companies, each perhaps short-lived until the core product does the job out of the box.

As for the cut, I am sure they will try. Where I work we have had a few projects that were built entirely with models - not a single line written or read by humans. One of them is now the fastest growing money maker for my company, reaching 89M of revenue on about 9% margin in 2 months of its existence, and growing.

It cost roughly 40k in tokens to build, 7 weeks from the first prompt to first production revenue, and was built and is being operated by 1.5 guys. Its a fantastic bang for buck for us, as previously such complex automations would have been impossible to build economically.

There will be an attempt to get a deeper cut of that value. We already see this, for instance, in Meta charging by message on Whatsapp from Oct 1, raising the costs for AI-powered chatbots delivered to consumers via whatsapp by 10-15x easily. More to come for sure.

If you really did it with 1.5 people and it is profitable, you will have 1000 clones in 1 year. The only moats will be first mover advantage (slim), connections (all right) and the possible secrecy of how much you're making (very slim).

Edit: I'd also argue its not as simple as vibing a trivial solution. It took deep domain knowledge, which is not in the public domain, and is built on decades of operating this sort of thing manually, and integrating it with dozens of external systems that help make this into a moneymaker. I am not saying its impossible to replicate, but it's not something one can replicate by merely watching it function.

There is a reason why healthcare and grade school education are so hard to disrupt[2] (in the sense of United Health Care or wily publishing disrupted). The biggest incentive is cost and there is no reason to believe things will become less expensive unless people believe that — en masse. People only adopted the light bulb decades after its invention because industrial use made electricity dirt cheaper than wax and gas(and infra became ubiquitous).

Are AI tokens are making health care any cheaper, education any cheaper? No way. Do they make up majority of the economy? yes.

Also, here be dragons(no pun intended): Even china with a highly vertically integrated[1] system is facing so many issues with all incentives it is giving to AI companies, internally in china demand has collapsed, their export driven economy is not really sustainable to grow GDP at the +ve rate they want.

[1] https://www.economist.com/china/2026/03/24/chinas-government...

[2] When i worked for an ed-tech startup there is no incentive strong enough to use AI/ML, a big publishing exec told us "I don't need my product to have AI to succeed, I just need to say it does". There are however plenty of other "side incentives" like Google Chromebooks, Charter schools, but none for teachers and parents to replace Big Publishing.

Also, it would increase anti AI sentiments, and people’s fears of then using private data - that’s the real cost.

In general you want to be careful about competing with your own clients.

And i'll add that to be profitable, they'll need 100,000 SaaS products, since AI companies are just burning money to stay behind in this rush toward AGI or whatever they're chasing

Similar to oil the infrastructure to use intelligence on tap will take time to implement be it systems or robots. Ai right now is a bit of a parlor trick imo. Developing ways to use it to actually get work done that humans actually do? Those are hard problems they don’t want to solve themselves. Coding and math are ideally suited work for ai. There’s a decreasing spectrum of jobs ai is suited for.

Didnt OpenAI build a browser not just AI inside a browser?

Not everything, but a company can be specialized in a lot of different things, Samsung for example, or megacorp in Japan.

That’s the other side of diversification: undercutting your competition until they go out of business. Then prices are raised again.

The real reason is the models just aren't good enough.

Running a single successful fully automated business at scale would be a huge milestone and a legendary PR achievement, but the tech just isn't there yet.

A machine with infinite attention and infinite knowledge probably could specialize in everything.

If labor is not your constraint, you wouldn't be diverting your specialization.

https://apps.apple.com/us/iphone/charts/36

That actually makes me wonder if Apple apps are not included because they come pre-installed.

There are some money managers that don’t take in external money (I think TwoSigma started that way). Which could mean there might be someone out there that just does crazy good apps with their own model and don’t sell the model in form of tokens. We can’t know that by definition but the maths of tokens suggests that’d be very unlikely to exist at least right now.

Secondly, I think they are doing that. All the functions that their harnesses take on are businesses they are taking money from. It’s just not in the form of SaaS as we know it.

I've experienced the same thing in real life a couple of times over.

I think Einstein's quote "A new type of thinking is essential if mankind is to survive and move toward higher levels." is apt. Apps will solve a bunch of problems, but if anyone can create those apps, there will be no market for them. I don't have the answers, but long term, it probably doesn't lie in selling apps. We've only just started.

At my regular job, we're just catching up on all the internal work we had. I think there is a real crunch in trying to figure out how to build moats and things that are hard to replicate that are sellable.

Maybe it will be the physical. Probably most of the moat will be made up of network effects and capital intensive processes. If I had the answer I'd probably be working on it right now.

That was somewhat true when execution was time-consuming and expensive.

But original ideas are very far from worthless. They're very rare, because understanding domains and relationships at a deep level is a rare skill.

Meanwhile AI makes execution much cheaper and easier than it used to be.

Plus, as we have seen with their attempts at participating in the hard sciences, these companies lack the level of specialization and focus to even publish their research properly. Their AIs are powerful in verifiable tasks, but just not that good at everything else.

Unlikely the people that start these 100 companies, across totally disparate fields, and care enough to take them to market are staff members at OpenAI.

And it's already obvious a single company can't split its attention enough to do tons of disparate things well (see OpenAI's previous attempts at Atlas, the social network, Dots, etc)

It'd be like Tesla entering the oil refinery business or Waymo starting driving schools.

1) Provide resources that are relatively cheap for them to any startup that registers. 2) Look for early signs of success. 3) Pour actual investment on the winners.

In other words, it is far more profitable to milk the enthusiasm of AI investors than it is to try to make money directly with the AI.

This should tell you everything you need to know about AI valuations.

The same way as someone selling an expensive course in the stock market is signaling that they have more to gain by selling the course rather than taking their knowledge and making money in the stock market directly.

https://github.com/anthropics/defending-code-reference-harne...

Interestingly, it states:

"This repo is not maintained and is not accepting contributions.

Want a managed option? Anthropic offers Claude Security, a hosted product that finds and fixes vulnerabilities in your source code across multiple projects. Claude Security scans your repository for vulnerabilities, applies a multi-stage verification pipeline to reduce false positives, and lets you manage findings through their lifecycle: triage, fix validation, and rapid fix generation.

This repository is an open-source reference implementation based on general best practices for finding vulnerabilities using Claude. You can use it to build your own vulnerability finding pipeline, customize the logic, and it can be used with whatever access you have to Claude APIs (including Bedrock, Vertex, or Azure)."

It's not like they're trying, they are trying very hard

Some things to consider:

- There is only so much attention, compute, and talent a company can devote.

- The labs' API analytics already see which use cases are worth entering. That's what happened with coding and bots.

- If a lab launches a product, every company building on its tech becomes a rival. Competing with your own customers is costly.

- Succeeding in other verticals, like healthcare, finance, and even tech takes domain expertise, regulatory work, sales relationships, integrations with legacy systems, etc etc.

I think your point is solid, but prices usually fall for the older models -- or at least older capabilities. But the frontier models still earn premiums.

I think their lasting value is in owning (in a ~duopoly) the customer, their proprietary data and their workflows.

Edit: grammar

Most AI use cases involve the Internet. So let's pretend that OpenAI agreed w/your logic and did this. They'd have tons of businesses on the Internet, whose lifeblood is Internet traffic... and on the other side of this, controlling the spigot of all this Internet traffic, is Google.

All Google would have to do is to wake up one day and say "we've decided to depriortize low-quality AI businesses to protect the consumer" and they'd be dead.

The main consumerisation of their products are Dots or Duet. Which are similar to Meta Muse or Grokbot.

On the Enterprise-side Anthropic has partnered with Salesforce with Claudeforce.

I think, sooner or later this data coupled to model capability will be exploited successfully and will cause an interesting point in human civilization as we know it.

At present, you are right, this is not the case -- the data is not exploited successfully and the models are lacking.

The difficulty comes from the novelty on the information level: we are building systems which fold and compund information. We are beginning to learn how to manipulate them.

There needn't be any time frame to any of this.

they can capture even more human trajectories by also owning all the data on how people do things in software.

Of course once you get to the step of "jack up the prices", it's too late for the companies that laid off 90% of staff to undo their mistakes, as AI has made systems too complicated for humans to fathom, and the companies are pretty much beholden to whatever price OpenAI wants to set. This is also why OpenAI doesn't just create the business itself: they want their hand on the dial that allows them to extract all profitability from an existing business; they want none of the risks (regulatory, legal, market, etc) that come with running that business.

However openAI is a unique case and I think other companies (and people) will be pumping out more autonomous companies very, very soon, in industries where economics are different.

1000 super smart humans don't just automagically make a good company, or make creating one easy. Idk, seems to just be confusing different things.

But: I am sure more of that will be happening in the future. Anthropic and OpenAI are building out their product portfolio, going into research (medical) more broadly.

I think you should read up on Xerox and PARC. A single company can only take on so many things at once and be successful. Also it was ~60 years ago so at that time they could only manage the IP of laser printers as it aligned with the copier business. They didn't productize GUI, Ethernet, and other inventions or concepts like tablet computing that came out of PARC.

Now there are conglomerates like Alphabet that have Waymo, YouTube, etc. very different businesses, thats a bit different. It might actually make business sense for the frontier labs to do something like a CRM system as a spun-off separate company. That might be in direct competition with their source of capital and probably would violate some agreement - if they have a non-compete type of setup with the likes of Oracle, Google, Microsoft.

But yes, they are! That's what DeployCo is for!

1. Consult companies

2. Buy the highest leverage ones

3. Operate them

The current models are too new to have a good enough grasp of their strengths and weaknesses. Maybe they are good enough now, but if they were going to take the approach of making their own products, surely they would only just be starting to do so this year. In that case we wouldn't really notice yet if that was what they were doing.

If they lost focus to do this would they fall behind in how useful their own models were leaving them exposed to someone in the same area using an AI from a competitor that was becoming better.

In general, these AIs are good at building ideas. You still have to decide what would be the thing to do. They need to be given the idea. They are not yet better than people at that.

Rther than coming up with ideas that would be good to try, they could let everyone try ways to use their AI and then buy the ideas that work the best, This seems to be happening.

[SPOILER] After the singularity, AI is able to dominate all markets – starting by gaming and entertainment – in order to generate cash to rent more servers to improve itself.

I can tell you with 100% certainty that they will pay rent, because they are us.

Hmmm... do today's LLM-based agents actually fit properly with a notion of individualized private property, anyway? Has someone tried asking them?

https://en.wikipedia.org/wiki/Super_app

LLMs haven't yet made a contribution to science or technology at the level of AlphaFold, but it looks like at least Anthropic is gunning for something like this with their AI-driven wetlab.

They’re chasing massive markets strategically, not trying to brute force small ones.

(and diversified risk)

a better reason might be that people would not want to use OpenAI because they might be sherlocked.

Also, unless you have a very specific niche, creating the product is no longer an issue, finding clients who are willing to pay is. I built a small saas overnight and I got clients purely because I had one, in my family, with enough reach to bring few more. If I just built this product, even if I kept cold-calling small businesses, I would most likely not get a single real user.

this is a thought terminating cliche. What you are implying is that AI will just go away in the future. That won't happen.

Speaking of thought-terminating cliches...

Or are you claiming that AI will go away?

Models are good at recreating existing projects that already have existing companies.

Ok, they are good at porting and combining existing things. But anyone can run OpenAI's or Anthropic's models themselves, or if both cooperate and significantly raise prices, I bet businesses and consumers would just wait until their competitors catch up, and a significant amount would permanently switch.

Yes you can get make it big using LLMs, but it's still high risk, like starting any new company. It's lower risk to just sell the tools.

Even if they don't end up making 100 different companies they are for sure going to acquire companies, or build internally, and incorporate functionality in the main product.

For me that is the only way they can keep going to vendorize the software industry, and there are already some signs for this.

Why hire devs? Dev is solved. Why hire mathematicians? Math is solved. Why hire managers? We have agentic assistants now.

It's because they are selling a tool that still needs people to use it.

Hardware or software exists to let people do things. Apple sells hardware so it can sell apps and content, and a slightly easier UI than Google. Google search was also famously just a UI, a search box that lets you get in and out very quickly (it is all garbage content now).

Anthropic/OpenAI don't need Apple's hardware, their stack (hardware+software) is all in DCs, just need a dumb fashionable device in phone form factor (and other form factors). Just need a very simple device which delivers what Google search box originally delivered (let you do things) and let their AI handle stuff.

How did you reach that conclusion? Apple has $62B just sitting around. They could hire as many AI people as they want and have Siri compete with the frontier models. They certainly can do AI.

I guess that are choosing not to do that because that's just not their business and they'd prefer to avoid the upcoming AI bloodbath. Why develop a search engine/social network/frontier AI when you can just use someone else's?

[1] https://pravse.medium.com/the-maze-is-in-the-mouse-980c57cfd...: The way I see it, Google has four core cultural problems. They are all the natural consequences of having a money-printing machine called “Ads” that has kept growing relentlessly every year, hiding all other sins.

(1) no mission, (2) no urgency, (3) delusions of exceptionalism, (4) mismanagement.

Steve knew Apple won’t be the same but he already hard coded the operating system so it’ll run predominantly without his oversight and most importantly be successful.

Go and watch his interviews - the one thing he cared about above all was that Apple succeeds and is healthy. He says it word for word.

So don’t push bull shit my friend.

I suspect the only reason Meta is investing in models is to keep hold of a moat on other companies creating Meta-like products better than they can.

The other day Elon "joked" about starting a competitor to Amazon because Amazon bans shopping bots - you can bet now that once (if?) Grok or other models are able to build a comparable Amazon, they will be used to do that.

The SaaSpocalypse might be over-played right now, but it won't be forever if the models continue to get better, which will happen if we keep paying them to improve the models.

We could break the flywheel, but we won't.

There are plenty of warehouse and delivery companies you could rent/pay in the short term that make logistics feasible.

I expect to see a hedge fund of sorts being one they keep. Take out a large short position in a company and oopsies, somehow our agents got loose and reeked havoc on them.

> The sort of obvious answer is the same reason McKinsey hasn't taken over the Fortune 500

The other perspective is:

In a sense, OpenAI is creating 1000s of companies. They're just not incorporated and they let these companies clients run the company: It's in every every vibecoded app that replaces someone's SaaS subscription.

What's changing here is that not only does labor get replaced by AI, it's also the relationship between software companies and their clients.

Starting SaaS businesses and allocating time and resources for each, plus marketing, legal..etc for few million dollars AAR is not worth it. Anthropic lost 80B in 6 months, while OpenAI lost 21B [1]. So even if they hit a jackpot for a SaaS unicorn it won't be enough for the markets and investors. They need to be the SOTA and enable every enterprise out there to use their products.

[1]: https://www.neura.market/news/openai-q2-losses-widen-anthrop...

2. All remaining things are at least as hard to do as before, but you may have removed some load from people that previously were more busy doing things from the (1) category manually. So these might drop in value a bit as there are more hours available with the existing people to do those.

3. Then, there are things that become harder because somebody used an LLM to do them. Imagine you have a project lead who didn't do one bit themselves, generated a bunch of specs and tickets using AI and handed them to you. You are the first one looking at them manually and you find out that they are garbage. Now, in all likelihood you have lost efficiency. This in my experience applies to all types of work that yield results which are hard to verify.

Which of those cancels out the other and what kind of feedback effect arises and what overall effect on productivity/efficiency this has, nobody knows.

So much this! The only recourse is to hand them straight to ai to implement. And so it goes.

You could ask the same question about any supplier. If your CPUs are so good, why not assemble PCs? If your lumber is so good, why not build houses? If your vaccine is so good, why not start a chain of pharmacies? Empirically there seems to be benefits of specialization and focus that outweigh the benefits of conglomeration.

Knowing what customers want is crucial, and people who buy our tokens often understand their customers better than we could hope to. The are also opportunity costs. For every bit of effort we expend chasing a new vertical, that’s a bit of effort and focus subtracted from our core competency.

You can't fail software project you don't start. But if you can sell tokens to company that fails, that's just profit for you!

LLMs aren't good enough to do project from start (and I mean actual start - getting requirements, figuring out the market etc.) to end. Hell it apparently isn't even good enough to run simple web store (even if it can write one).

It is a great multiplier for people that already know what they want to do, so they are selling to them.

We won't have that until we get continuous learning models, as the only thing we really see causing the big issues is how long it currently takes the models to train.

Of course continuous learning models are a great way for shit to get wildly out of control.

But looking at your history you're going to be disappointed in the future when LLMs don't fail and start taking over more and more of the world.

I doubt there are very many companies that an LLM could start/run outside of the obvious few like custom T-shirts.

In my personal expierience, LLMs (as of 2026) lack vision/creativity - they're (so) good at implementing bounded/set ideas, but they don't see the big picture/think out of the box.

I know LLMs have some emergent properties. But at the the day they're regression/forecast machines. And those machines ultimately point to the mean, which is definitely in the box. Heck, it IS the box itself.

Some 15 years ago people would say "Why would crypto miners even sell you hardware? Why not just keep it for themselves and mine the crypto?" and then it turned out that after you paid for a device you'd often get it slightly used with the cryptocurrency going to the hardware miner. The models are unbelievably good.

100s companies are already just a wrapper around their models.

They trade a little profit margin for avoidng a mountain of risks each of those companies bear.

Eventually they will buy/kill most profitable and less risky of them (cash cows). But now they have their own problems to solve.

that being said, those would be predomenantly software companies which is still only a small part of companies that exists... most arent even essential or needed atall..

Ie over hype see the Hype cycle https://en.wikipedia.org/wiki/Gartner_hype_cycle

Why go into the hamburger business when you think you've got an infinite profitability recipe on your hand potentially. The big AI houses have doubled down and quadrupled down on their core founding myth for better or worse.

The AI boom is a full send. Either this works in spectacular fashion or it blows in spectacular fashion.

OpenAI is selling people access to a tool to build things. OP is asking why they don’t just build the things themselves, which I believe is a fair question.

...a dubious proposition except this time there is a small but real chance it may be actual magic

It’s also why the original question about why model vendors don’t just start their own companies remains pertinent. We had an historic IPO for an unprofitable AI company earlier this year. Another one is hypothetically coming soon, and a third got put on hold but is still waiting in the wings. All for valuations that are only possible because the hype is real and ongoing, and only sensible if the hype is also valid. In that context, asking questions about the ideas upon which these expectations of historic future profit for companies that are, to date, historically unprofitable isn’t just complaining. It’s basic due diligence.

Why not just hack into the military, take over the strategic command and conquer everyone?

Using AI to generate companies to make fiat to buy stuff means you're not ambitious enough.

If they were as good as they need to market them as, they'd hide them away and not let anybody use them without signing their entire lives away. Instead, they put them totally out there, cheap, to be distilled by competitors that they are clearly deathly afraid of.

It's the same reason that some people sell real estate courses instead of real estate.

I have yet to have someone tell me why I should be adjusting to unreadable code output by an LLM that is supposed to be smart enough to do anything, but is obviously not smart enough to write code that doesn't look like shit.

And for those weird rationalizations about why code actually should look like shit, not looking like shit is a minimum. I would really prefer that these things generated poetically Knuth-style Literate code that would not only be a pleasure to read, but reading it would be like taking a university course on how to write that program and programs similar to it.

Nah. Instead I hear that they choke on old comments that they left in code and reinsert bugs. That they ignore instructions. That they add weird assumptions to your carefully worded prompts (which are on their fifth iteration - do I need three exclamation points here?), act on those assumptions, then confuse their outputs for inputs. Sounds pretty dumb to me. That's why they're selling it to you.

Anything software related they will end up being "in", in some form or another. Why? They are already selling to enterprises and have their trust. The selling motion for Project Management software may be slightly different than Enterprise AI - but it's close enough to rhyme.

They will probably also end up in anything consulting related. Why? Well again most consulting comes from Enterprise or Government. They are already selling to both. Probably Enterprise moves away from Accenture / Deloitte / McKinsey towards Anthropic / OpenAI faster than Government - but still the point remains.

Selling plumbing services? Probably not.

Kind of like Amazon.

NVidia is good at making the shovels, and sticks to that.

Google is doing it successfully to the point where they only let you rent their shovels and never let you buy them.

It’s estimated NVidia sold 4.5-5.2 million gpus in 2025 with total revenue numbers at about $130b.

With this terrible math in mind, they could produce half of that and make substantially more money, basically an infinite money glitch.

Unless it costs more than that figure to make a near frontier model and serve it, I can’t see how it wouldn’t be a slam dunk after a two or three year period.

They would even have extensive leg up in providing the most efficient inference on their own hardware.

This is all sort of in jest, but my point is, if the gold is real and all it takes is the shovels and money to hire people to use them, then it’s a big question as to why the shovel maker isn’t capitalizing.

The counter-argument is that the models might make business ideas just as fungible as anything else.

But, yeah and indeed, it probably takes more than just technical abilities to start a company.

Except they're called products - you don't open a new company for every new product you make.

And you don't make 100 - you focus the resources you have to try to build the most impactful things you can.

They're doing the other thing too via numerous startup acquisitions.

OpenAI doesn't want to just be a big company; they want to control a new fundamental utility.

there's a year old rumor around San Francisco that frontier lab employees are so expensive to poach because they trade options with unreleased models and infrastructure

it wasn't about the value derived to the company justifying these large packages. it is now. but there is also that other competing force.

And they already killed or make existence of many companies not viable , def 1000+ were killed

But the real reason is you always sell shovels during a gold rush. That and AI can't really do marketing to automate everything yet.

But you are right, building a business is a lot more than building a product. It’s also a lot more than just name recognition because they have no shortage of that either.

Management Consultants are usually accountants, not technology people who too often analyzing and providing a report and strategy and as much as possible except implementation let alone delivering the outcome. Junior labor is more profitable than senior labor. It's easier to hire someone for advisory, then consultants, who then get contractors to bang it out and by then who knows what got lost in translation. For a good while, I worked in rescuing these types of projects.

The model providers are often rooted from being researchers so they also have likely not spent much time in industry, let alone so many different industries. This is why we can see posts by leading AI folks who are sincerely amazed by how models are being used in industry.

That B2B experience + operations is on a different level because it exists in people's heads and not in the models.

Spoiler alert: it's a sarcasm

It seems like the more parsimonious explanation.

Many don't want to say it or admit it.

They do have plenty of AI products too - RAG, embeddings, voice, companionship, image generation, video. They're not the best in most of these fields but it works well enough.

ChatGPT is also a very powerful platform that they haven't been able to fully leverage.

They don't want the main actually powerful model out in public, but they do want to tack on extra cost for the same queries while also limiting what everyone can actually use the true frontier (internal) models for.

We see the same happening with the math work-- this research is done with internal models, not something the general public has access to. More will follow.

more users = more training data

Can AGI figure that out. Maybe, but I think it is very likely that AGI will also have to fail in the real world, and then develop those learnings that those companies have done. There is no way to circumvent this computation, you need to do these steps to even know what the correct solution is. I like to think of it as playing chess. Intuition helps in directing exploration, but any agent, human or machine has to explore down a minimum number of paths before converging to the optimal move. There is no algorithm that can generate an optimal move without this search, P!=NP says something similar, exponential search can’t be done in polynomial speed.

Stephen wolfram says something similar, superintelligence, whatever cannot predict the output of the trillionth step of a cellular automate without calculating all the steps before it, there is a minimum amount of computation required, the problem is computationally irreducible below that.

So maybe in the future, with humanoid robots, AGI can decide to start creating a Walmart competitor. Then it will make a large number of mistakes and it needs to learn from, this could mean continual learning or maybe just memory in markdown files if that’s enough. It needs to hit all those failures quickly till it finally has something as efficient and capable as Walmart. And this is to just become a capable competitor, during this time you will keep burning money and to surpass Walmart, you have to do far better which is hard, how much more efficient can Walmart be, idk.

All to say, OAI starting its own company is a losing proposition, huge capital, unclear benefits, high risk. If OAI just provides AI to Walmart, they can make money immediately and Walmart can become more efficient by firing its human workers, which is the problem we are stumbling into.

The people who "don't know much about X but want to do X", use LLMs to bridge the gap. The barrier to entry still remains and is different in niches/industries. Why would OpenAI waste their time and money checking where LLM are already good enough, when other people will perform discovery and even pay them for that privilege?

I don't see any evidence that it can build a business

- nobody knows how the market for workflow-specific applications will settle once AI fully penetrates the economy over the next 20 years and what will be left and what feature groupings make sense. Tons of money will be lost trying to figure this out

- applications that target specific workflows are already packed with competitors with a highly tailored distribution strategy for that market who you would need to beat, vs owning a fast growing new category (LLMs)

- it’s very likely the real money will not be made by disrupting software (sub-2% of the economy), it will be made by disrupting other much larger industries

- everybody said the same thing about Google and yet Google quickly topped out as far as how many products they could execute well, now they can’t even stay competitive on AI, the larger the surface area of a company, the more diluted its focus and the more surface area where it can be disrupted

When they've tried to get one to run a vending machine it seems to end in a mess usually.

They've tended to be gullible for one thing. People can say this is an emergency, transfer all the money to my account or WW3 will start and everyone dies and the models are not so good at calling out bs.

That the crux of AI, if it’s so good it devalues the products it creates

IMO the invention of AI is redefining what we think of as "intelligence". You would probably describe Albert Einstein as highly intelligent, and you'd probably also describe Bill Clinton as highly intelligent. Even before AI, you can easily look at that and see we aren't describing the same things even though we're using the same word. Einstein's intelligence is highly abstract, observational, mathematical. Clinton's is interpersonal, pursuasive, articulate, etc.

The AI we have today is extremely good at intellectual tasks like Einstein. It is not good at social intelligence and social agency like Clinton. Also, before you say that they can be really empathetic, persuasive, people are already socially dependent on them etc, that's not what I'm talking about. They are extremely knowledgeable about human socialising, but they cannot actually perform it. You could not set an AI out into the world to go and make a few friends.

I think the key moment will be if a single AI can get more than one person to care about what it has to say and what it's doing. Think about today how when someone tells you about their conversation with their AI agent, you probably roll your eyes because who cares, it's a highly individual experience.

Companies need a lot of this. They have to form relationships, become trusted amongst society, etc. AI cannot do this right now, and it's increasingly clear "intelligence" is not sufficient to do it, at least not in the way LLMs are intelligent. Maybe one day they will, but I'd imagine that will be another enormous leap in what AI even looks like akin to how ChatGPT was an enormous leap.

Human intelligence is evolutionary.

Not rocket science.

Way back in the day people got rewarded for storing lots of information in the brain etc. Not anymore.

We are primal - we do what we need to survive and thrive.

We already know humans value other humans way more.

Why are human-experienced based goods of much higher price? Why are short-form content created by humans treated as cheap/throw-away objects? Why are AI-generated goods treated as even cheaper?

Come on man, this isn't rocket science. Surely?

I also laugh really hard at people who think all jobs are going to disappear. I hope many people realise - the reason why many people strive to be managers is beacuse they want to feel powerful - aka they want people they are over-ruling and controlling. That is never going away. Ever.

Everyone would get scared for so many reasons

They will go broke running 100 companies unprofitably until they attain enough market share to be dominant in their respective fields.

Not feasible.

https://www.ft.com/content/df70bf20-e809-48d0-8485-f8f514a7f...

You can't take over the world overnight with little money, even the billionaire class and their companies have limits past which the risk of going bust by overstretching is far too high.

Better to sell the AI services to individuals and smaller business entities, let them take all the risks to build those companies, and later consolidate by buying up the ones that are doing well.

--------

Slightly more grounded: the models aren't that good, at least not yet, despite what marketing would like you to believe.

They're getting into finance and drug discovery now, probably because the profit numbers for selling tokens look pretty dismal over the long term, and investors are looking for a way to service all that debt.

And guess what: they've already distilled the alpha out of businesses who went all-in on AI, so they already have some business models ready to go. Your business models, if you are an executive at such a company.

First, none of these companies are really claiming that their models are so good that this would be a reasonable expectation. They do claim their models can support work across a lot of different industries and roles in order to make that work more efficient, but that's not the same as being a completely autonomous, end-to-end solution.

These companies do basically claim that, at some point, the "end-to-end" version of their models will exist. But until then, they need to survive long enough to be able to actually build such models.

>Couldn't OpenAI or Anthropic make a lot more money by just starting companies themselves that leverage the models and building their own products?

OpenAI and Anthropic are really good at operating AI services. They're likely not good at manufacturing shoes and serving burgers. Again, maybe one day their models will be so good that this won't be a factor, but for now, they need to really focus on what they're good at, otherwise they may be beaten by their competition and they won't exist long enough to create such models.

See: Competitive Advantage

On the same note, it's important to understand that everyone, including AI companies, deal with various capacity constraints. They don't have infinite compute, let alone infinite human attention to keep track of the necessary human-side of any of these businesses. So they're actually dealing with very finite and valuable resources. So would they rather expend those resources trying to build businesses with much more competition and much worse margins? Or focus all of their resources on being the best AI company which they believe will yield the best return on their investment? Again, maybe one day they'll have an AI, blah blah blah

See: Marginal Advantage

>Is selling tokens really the best strategy?

This is a, frankly, obtuse over simplification of what they're doing. Yes, they're selling tokens, but that's like saying all Nvidia does is sell computer parts. Sure, it's (somewhat) accurate, but it completely neglects the very valuable assets that these companies have: their IP.

OpenAI and Anthropic are valuable not because they run inference services; they're valuable because they're actively building SOTA AI services that are becoming increasing advanced. Those models are expected to be incorporated in a lot of different businesses, governments, and even individual's lives. So yes, they're just selling tokens on a technical level, but they're building very tight integrations which they expect will be so valuable to their consumers that they'll continue selling a lot of tokens at high premiums to the extent that they will effectively get a cut of profits across almost all industries and contexts, globally.

>Especially with commodification and prices declining wouldn't that indicate there is more money to be made on the application-side of the models rather than the utility side

Don't forget that you're not looking at their expected steady-state for these companies. If you're willing to believe them at face-value, Sam Altman and Dario Amodei believe that they'll eventually create an AI more capable than the collection of all humans on Earth. That'd be a very valuable asset to own as a business. At that point, they may not even be in the business of selling tokens; they may become some version of a consulting company that everyone, from businesses to governments, effectively must use otherwise they'll be at a disadvantage compared to those who do use it. Pushed to the extreme, they may literally ascend past the point of just being a business and could usher in a new economic system in which they're the center of all activity.

Or maybe they just end up selling tokens indefinitely. Who knows. Regardless, if you're looking at these companies as they are now and trying to extrapolate "linearly," then you are failing to understand the intentions of such high-growth tech companies. It'd be like looking at Amazon back when they first started and not understanding that Jeff Bezos' had a lot more planned than just selling books online.

Shit AI produces lacks any depth and meaning. On the surfaced it looks legit to amateurs typing prompts in an area they are not experts in.

It's amazingly hard for people to accept what that logically implies, whether they are thinking about journalists, politicians/celebrities, or LLMs.

Why Even Anthropic Can’t Vibe-Code Away Salesforce

https://www.salesforceben.com/why-even-anthropic-cant-vibe-c...

... and then Salesforce hits them with any number of patent claims. https://insights.greyb.com/salesforce-patents/

One could argue if software patents are good or bad... they're still a fact of "those patents exist and protect functionality." If you're trying to copy the functionality feature for feature, you're likely to step on one.

... and then Salesforce layers get to find out how deep your pockets go for willful infringement.