Something big is happening; but it doesn’t add up

This week, multiple people sent me a link to Matt Shumer’s post that Something Big Is Happening. In the post, he sounds the alarm that AI is coming for most white-collar jobs. It’s definitely worth reading:

If you extend the trend (and it’s held for years with no sign of flattening) we’re looking at AI that can work independently for days within the next year. Weeks within two. Month-long projects within three.

[Dario] Amodei [(CEO of AI company Anthropic)] has said that AI models “substantially smarter than almost all humans at almost all tasks” are on track for 2026 or 2027.

Let that land for a second. If AI is smarter than most PhDs, do you really think it can’t do most office jobs?

Think about what that means for your work.

[…]

I think the honest answer is that nothing that can be done on a computer is safe in the medium term. If your job happens on a screen (if the core of what you do is reading, writing, analyzing, deciding, communicating through a keyboard) then AI is coming for significant parts of it. The timeline isn’t “someday.” It’s already started.

His advice, which is to start paying for Claude or ChatGPT and using it to do more than search for information, is not bad guidance, but it seems disproportionate to the threat that he covers in the first part of the article. There is, Shumer implies, a short window of time in which you can be “the most valuable person in the room” by supercharging your productivity with these tools. Most people won’t do this, of course. He argues — correctly, I think — that if you spend an hour a day experimenting with the software you will gain a significant advantage over your peers:

One hour a day, every day. If you do this for the next six months, you will understand what’s coming better than 99% of the people around you. That’s not an exaggeration. Almost nobody is doing this right now. The bar is on the floor.

The terrifying stuff is where he talks about needing to “get your financial house in order”, building up savings and avoiding “taking on new debt that assumes your current income is guaranteed”, presumably because of the coming wave of job losses.

However, there’s something about where all this is headed at a macroeconomic level that doesn’t add up for me. The massive investment in the technology, specifically processors and data centres for both AI training and inference, is largely funded by debt. Companies like OpenAI and Anthropic are losing money. Fortune reported in November that:

OpenAI is plotting a dramatic arc toward profitability through the end of the decade, but that growing won’t come without some pain. The company reportedly expects to rack up massive annual losses each year, including roughly $74 billion in operating losses in 2028 alone, then pivot to meaningful profits by 2030, according to financial documents obtained by The Wall Street Journal.​

In December, they wrote about Anthropic:

To be sure, Anthropic is nowhere near profitable today: It was pacing to end 2025 having consumed $2.8 billion more cash than it took in, according to recent news accounts citing forecasts provided to investors. But the company is also on track to break even in 2028, according to those projections—two years ahead of OpenAI.

If a bunch of us pay $20/month for premium access, with a tiny few paying $200/month for the highest tier, I don’t see how this goes anywhere near covering the costs. (Yes, I realise that there are Enterprise plans and paid tiers for API access, but the point still stands.) If this tsunami of AI turns up and eventually takes on most white-collar jobs, making loads of us redundant, there will be even fewer of us able to pay. OpenAI are testing adverts in ChatGPT and they may help, but I don’t see how displaying them to people who are asking the computer to do something and then coming back days later to see what it did will fund the operation either. Microsoft Copilot, which leverages AI technology from OpenAI and Anthropic, is paid for on a per-seat basis. Job losses would mean fewer seats and less revenue. Maybe the people that are left with jobs will find the cost of their licences increasing in response.

Even Google, which can use its vast ad revenue to fund Gemini, its own AI product, and keep it ad-free, would likely have its revenues go down if there is a seismic economic downturn. If people can’t afford things because they are unemployed, what’s the point of advertising something to them?

Maybe I’m looking for logic that doesn’t exist. Like Bill Joy said in his Wired magazine article 26 years ago¹, people will invent technology and do things without understanding the consequences, just because they can:

Perhaps it is always hard to see the bigger impact while you are in the vortex of a change. Failing to understand the consequences of our inventions while we are in the rapture of discovery and innovation seems to be a common fault of scientists and technologists; we have long been driven by the overarching desire to know that is the nature of science’s quest, not stopping to notice that the progress to newer and more powerful technologies can take on a life of its own.

At this point, while we don’t know exactly where we’re going, I can’t disagree with Shumer’s advice: start using the tools in more advanced ways, figuring out what does and doesn’t work in your context. As for the bigger picture, I guess we’ll just have to cross our fingers and hope?

1. Joy’s article is something that I’ve come back to more than anything else I’ve read in my life. ↩