Except... Few are actually using it. So what, exactly, is the value in MCP?
Especially that there are simple ways for anyone to spin their own MCP based on standards like OAS. I talk to dozens of new clients in a given week. Our product should attract users who want MCP. And in the last month only one conversation actively asked us if we had an MCP server. Surprised, I asked about use case and the response was as I'd expect: "No specific use case, we're just playing around with it". Seems to be pretty standard for AI conversations these days.
I don't think this is dissimilar to the US to be honest. We just happen to have inflated and questionable revenue being reported at levels that gloss over the true underpinnings of the economy. I think this may be a leading indicator.
I really liked how Cory Doctorow framed what Canada could be doing with respect to digital sovereignty in his book ("Enshittification"). Doctorow argues that if Canada repealed its anti-circumvention law (Bill C-11), Canadian companies could legally jailbreak American tech products (John Deere tractors, the Apple App Store, etc) and sell those fixes worldwide, turning the right to tinker into a massive export industry and a form of digital sovereignty.
(Personally I would love to see Sam and Dario credibly walk back their jobs doomerism talk, but I don't think the facts match the headline in this case.)
But you also don't know. So, it seems like a contradiction.
Also, for fun, I asked Opus 4.8 to confirm or deny this... Full text:
"""
Confirmed. The quote is accurate.
Altman said it on May 26, 2026, during a virtual interview at a Commonwealth Bank of Australia (CBA) technology conference in Sydney, in conversation with CBA CEO Matt Comyn. He was speaking remotely to the conference about AI's impact on jobs. (International Business Times)
The fuller context: Altman was walking back his own earlier predictions about AI-driven job loss. He framed the remark by saying he was "delighted to be wrong about this," then added that he thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened. (eWEEK) He continued: "I now think I understand more about why it hasn't, and I'm obviously grateful, but that is an area where my intuitions were just off." (Time)
His explanation for the miss centered on human interaction. He argued that human interaction remains an irreplaceable component of many jobs and that even as AI handles routine tasks, there is a "human part" of employment that can't be substituted. (International Business Times) He mentioned that after trying to delegate his own Slack and email responses to AI and then returning to doing them manually, he concluded "we really do care about our interactions with people." (Breitbart)
A few contextual points worth flagging:
The reversal is notable because Altman had said the opposite roughly a year earlier. On the Uncapped podcast with his brother Jack in June 2025, he had predicted many jobs would disappear due to AI, though he expected new opportunities to emerge. (Breitbart)
The remark functions partly as a contrast with rivals. It lands as an implicit rebuke of Anthropic CEO Dario Amodei, who has warned that AI could eliminate up to 50% of entry-level white-collar jobs within five years and push unemployment to 10–20%. (OfficeChai)
Some coverage notes the timing. Several outlets point out the softened messaging comes as both OpenAI and Anthropic reportedly prepare for IPOs, each at valuations around $1 trillion (Breitbart) — which is context certain reporters are offering as possible motivation, not something Altman himself stated.
It's also worth a caveat on sourcing: the quote itself is consistent across many outlets (Time, Reuters via Fast Company, IBTimes, eWeek), so I'm confident in its accuracy and attribution. Some of the surrounding framing (the IPO angle, the Amodei "rebuke" characterization) reflects editorial interpretation rather than Altman's words.
"""
Wow, that's a bad answer from Claude. I looks like it cited half a dozen publications who all pretty much just copied and pasted the same tiny fragment of quote from each other.
Note that I'm not denying he said those things, just that I don't think there's a credible path from a couple of sentences in that one interview on Tuesday to headlines like "Sam Altman and Dario Amodei are both walking back AI jobs apocalypse predictions".
(It's amusing to me how many people accuse me of "hype" while simultaneously being quite happy to consume hyped headlines that match their priors.)
No. This is a ridiculous take on the recklessness Altman and Amodei have stated. Both men flat out lied, kept lying and continue to lie about their numbers and capabilities. They have literally destroyed markets that aren't coming back anytime soon (consumer hardware). They should be held accountable, not let off the hook like "Oh well, they fucked our economy long term. Guess they're just human."
Literally fuck them and an oversimplification like this.
I'm not going to watch a 3 hour video, but custom PCs are a tiny fraction of the consumer hardware market. Also, the custom PC market is struggling partly due to people overstating the price of RAM. It's not like the cost of building a PC has doubled. RAM and SSDs were typically some of the least expensive components in a complete build. Looking at prebuilt PCs from iBuyPower, a pre-built that would have cost $2000 last July now costs a whopping $2100 with the same components.
I think this is a big component, but also context. A large factor in any model being able to handle complexity comes down to context length.
I think multiple SLMs driven by an orchestration frameworks (harness or otherwise) will ultimately displace LLMs. Right now we're in the era of diminishing returns with respect to LLM gains. Moving the needle percentages doesn't excite as many people anymore and with "reasoning" capabilities there's no reason why small distributed models can't be run more efficiently, especially if/when we start to see gains in modularized context management solutions.
It's hard to know for sure. There are good information theoretic reasons to suspect that general models will always be better than smaller expert models, but maybe a MoE can claw some performance back, albeit with redundant computation. The properties of conditional entropy, for instance, always favor more generality. This assumes that the harness isn't a factor, or is at least equivalent across different models.
The difference is that they had room to charge more of their customers and pay less to their workers. The AI industry doesn't have both sides to play at this point. Training and inference are getting more expensive and if you take on the high prices now you're just floating yourself further downstream from profitability long term (which does not look viable for any of them currently).
Here's a source from 2019 that says: "By 2023, the number of knowledge workers in the world will increase to 1.14 billion, with more than four-fifths of that growth coming from the emerging world."
To add an actual source to this thread, a brief paper by researchers at the International Labour Organization (ILO) states that for knowledge workers globally "... there
are between 644 and 997 million jobs, which represents
between 19.6 per cent and 30.4 per cent of global
employment respectively." [1]
[1]: Berg, Janine and Gmyrek, Pawel, Automation Hits the Knowledge Worker: ChatGPT and the Future of Work (April 21, 2023). UN Multi-Stakeholder Forum on Science, Technology and Innovation for the SDGs (STI Forum) 2023, Available at SSRN: https://ssrn.com/abstract=4458221
Globally, sure. The assumption here is all users are on the same economic footing, they are not. Only about a 1/3rd (at most) of that count can afford $1000+ monthly, and even then that is wildly out of line with what most will.
I googled "number of knowledge workers worldwide" and read the top results. If you read it as I was confident in a billion I apologize, Im just trying to get an accurate count. What numbers do you have now and where did you find them?
That's not the TAM of 1B knowledge workers globally. If that were the case many industries would have a 2-3x target market.
To simplify break that 1B up into 3 levels of purchasing:
1) High-tier (US, Western EU, ANZ, Japan, South Korea, Singapore, UAE, etc) - 200-250M knowledge workers.
2) Mid-tier (Eastern EU, Latin America, urban China, India tech sector, etc) - 300-400M
3) Low-tier (Rest of the world) - 300-400M
Low-tier users are mostly free tier or heavily subsidized pricing.
Mid-tier are going to account for USD sub-$100 tiers. Probably averaging less than $50/seat.
High-tier are who you are assuming is the 1B. Users are not equal in that knowledge worker count, so there aren't 1B knowledge workers to charge money.
And when you consider Low-tier users a majority of those are free users which need to be subsidized by the High-tier users. So either free tiers get much more restrictive or the providers lose additional training data. A bulk of Low-tier users cost money and provide little to no revenue.
Edit:
And think about Mid-tier and Low-tier for 5 seconds. Why would they pay Anthropic or OAI when they get get 100x+ inference from DeepSeek or Xiaomi? Mid-tier may be the only area that is willing to spend money on a US provider, but I would wager significantly on the fact that users in the Low-tier almost universally do not care.
Thank you. So with these numbers it seems like half a billion subscriptions at $500/yr is on the table. Obviously theres going to be competition in this market and self hosting cheap models may become the dominant use case. Assuming the labs are able to get most of the market though, the market size is something like a quarter trillion a year within the next decade. It's hard for me to imagine the whole sector failing if that happens.
I do think free accounts are going to end pretty soon, and some of the workers in your tier 3 will pay, but even without them this seems like a pretty healthy market size. I also wouldnt be surprised if mid tier workers are able to afford the $1000/yr vs $500. I use yearly rates because I find it easier to compare them to GDP/salary numbers
I mean, sure. Assume all you want but to guess that the entirety of High-tier plus almost all of the Mid-Tier will spend, on average $500 per annum is bonkers.
I believe we've started to see the top of what individuals and businesses are willing to pay for the current model capabilities. We are nowhere near AGI and models are really only providing significant value in niche markets currently (programming and cybersecurity). And just like SaaS the enterprise has the option to buy hardware and leverage their own models at will which can potentially offset costs and TAM as well. I have talked to a number of large financial corporations in the last 6 months and most have internal initiatives. The same applies in the healthcare vertical.
$250B per annum with AI? That's 20% of global software spend now. Sure, that's possible but that assumes current market prices hold. What if inference ends up normalizing between DeepSeek/Xiaomi & Anthropic/OAI? There's 50% of your revenue and with current costs for inference and training in the US at astronomical levels the US AI industry could also very well be setup to implode overnight.
Lastly I don't believe free can go away anytime soon because it can't. As soon as Anthropic and OAI remove that option those users will move to whatever is. For most of those users it's not a luxury to choose, it is the only option.
The financial engineering occuring right now is something I don't doubt will be text book lessons of the future. We've seen it before and I believe Peter Sorkin when he says that we will see a crash of this bubble, it's just a matter of how catastrophic it ends up being.
I agree that a lot has to go right for the AI labs for it to work out for them, I just dont think it's already over like the top comment in this thread seems to. MSFT makes $55 billion on their office products, the AI labs can use a similar strategy I think. I think AI assistants will be more indispensable than office products within a decade. Hard for me to imagine doing office work without an assistant a few years from now, but maybe I need a better imagination.
My 76 old Dad loves checking the TOTP on his phone and asking me to verbally read it out when I need him to accept a 2FA push notification to let me log into his bank or government accounts so I can do something for him.
He says it “feels like 007 stuff.” “AI will never trick me!”
We also have a duress code word, listed in the notes of that KeePass(ium) entry with the TOTP.
You have no recourse in the US, either. Trust no one is the only path given all of the training data is stolen in the first place.
It will come to light that one or many of the Frontier providers held the data, changed ToS and trained later minimally. But I think they just don't care and will train regardless. None of them abide by any level of ethics that would actually prevent them from leveraging an opportunity.
"Everyone is building this!"
Except... Few are actually using it. So what, exactly, is the value in MCP?
Especially that there are simple ways for anyone to spin their own MCP based on standards like OAS. I talk to dozens of new clients in a given week. Our product should attract users who want MCP. And in the last month only one conversation actively asked us if we had an MCP server. Surprised, I asked about use case and the response was as I'd expect: "No specific use case, we're just playing around with it". Seems to be pretty standard for AI conversations these days.
reply