From an economic standpoint, for these companies to see a return on their investment, won't they need to replace jobs? It will be challenging to recoup investments by charging regular users in a post-DeepSeek era. While I don't support job losses, aren't they the expected outcome?

AIs replacing jobs is not the only way those companies can see a return on investment, it's not necessarily zero sum. If the additional productivity given by AI unlocks additional possibilities of endeavor, jobs might stay, just change.

Say idk, we add additional regulatory requirements for apps, so even though developers with an AI are more powerful (let's just assume this for a moment), they might still need to solve more tasks than before.

Kind of how oil prices influence whether it makes sense to extract it from some specific reservoir: if better technology makes it cheaper to extract oil, those reservoirs will be tapped at lower oil prices too, leading to more oil being extracted in total.

When it comes to the valuations of these AI companies, they certainly have valuations that are very high compared to their earnings. It doesn't necessarily mean though that replacement of jobs is priced in.

But yeah, once AI is capable enough to do all tasks humans do in employment, there will be no need to employ any humans at all for any task whatsoever. At that point, many bets are off how it will hit the economy. Modelling that is quite difficult.

> once AI is capable enough to do all tasks humans do in employment, there will be no need to employ any humans at all for any task whatsoever

AI has no skin, you can't shame it, fire it, jail it. In all critical tasks, where we take risk on life, health, money, investment or resources spent we need that accountability.

Humans, besides being consequence sinks, are also task originators and participate in task iteration by providing feedback and constraints. Those come from the context of information that is personal and cannot be owned by AI providers.

So, even though AI might do the work, humans spark it, maintain/guide it, and in the end receive the good or bad outcomes and pay the cost. There are as many unique contexts as people, contextual embeddedness cannot be owned by others.

>But yeah, once AI is capable enough to do all tasks humans do in employment,

Also at this point the current ideas of competition go wonky.

In theory most companies in the same industry should homogenize at a maxima which leads to rapid consolidation. Lots of individual people think they'll be able to compete because they 'also have robots', but this seems unlikely to me except in the case of some boutique products. Those companies with the most data and the cheapest energy costs will win out.

Forget about that.

Lets focus on the tech firms that produce software.

Two things should happen if AI proliferates into software development:

1) Increasing top line - due to more projects being taken by enabling labour to be more productive 2) Operating margin increasing - due to labour input declining and taking more cost-reduction projects

If those 2 things dont occur - the AI investment was a waste of money from a financial perspective. And this is before I even discount the cash flows by the cost of capital of these high-risk projects (high discount rate).

At some point everyone will be analysed in this manner. Only Nvidia is winning as it stands, ironically, not because of LLMs. But rather because they sell the hardware that LLMs operate on.

I would also add that many (most?) companies/entities do not sell software but have large IT departments that could write software for internal consumption. Think Exxon, BP, Caterpillar, Airlines, Gov Labs/agencies, DOD, etc...

Internally, they could actually write 1,000X more software and it will be absorbed by internal customers. They will buy less packaged software from tech firms (unless it's infrastructure), internally they could keep the same headcount or more, as AI allows them to write more software.

If we need those two things to occur, then any money spent not on salesmen and golf tee times is a waste. There are so many outside factors that aren't product features that make or break a company and a product's success and while we, as software developers who work on the product and its features want to believe things like code quality are relevant, but they're not as important as we'd like to think.

That only gets you an expected net present value. Looking at the variance and quartiles is way scarier.

The hucksters will tell you the variance in the cashflows is exactly why they are pursuing AI - real options.