For the last many months, I’ve been casually reading and thinking about AI’s net impact on the job market. Studies from BCG and OpenAI show that LLMs are already making white-collar workers around 30% more productive.

History is a testament to how despite fears identical to what we see today, generational technological revolutions like machines and personal computers have only reshuffled employees into new jobs while massively increasing total GDP. That said, I’ve always reasoned that there is something qualitatively different about a machine that can replace human judgment. Because of this, I’ve been fairly pro-automation-concern until recently.

If you’re at this stage, I get it.

This post is to document what has changed for me, mostly because I haven’t seen anyone else articulate these ideas similarly yet.

In this short piece, I’m going to be categorizing jobs in two ways.

• Demand Constrained v Non Constrained jobs

• Jobs in Manufacturing v Service v Software

To be clear: I’m not fighting the claim that many jobs will be lost. Just that automation will risk the survival of a healthy job market, and that the obvious solution is something approximating Universal Basic Income.

The Argument

As Americans move increasingly into professional information work (we’re around 60% today), autonomous AI is going to automate away many, many jobs. I would be comfortable wagering it will eventually automate most jobs we have today.

But in a market with free flow of information and capital, there isn’t a fixed number of jobs we’re all fighting for.

On the firm level, as Aswath Damodaron of NYU puts it - if everyone’s got AI, no one does. i.e. if all firms have access to the same kinds of AI models at similar rates and similar external or internal datasets, there is no long-term price or profit advantage to be found in implementing LLMs.

That is, as long as you agree with one additional qualifier: virtually every firm of virtually every size will always need some human management. That there isn’t a future world where digital nomads in Bali have a personal agent building and profitably iterating on 1000s of different online businesses simultaneously & so passively as to need zero human oversight.

I agree with this qualifier based mainly on strong intuition, but also because a world where no human oversight is needed in every sector is a world where we’ve effectively lost the idea of working for a wage. For obvious reasons, that is likely net positive too.

Assuming my qualifier is correct though, here’s how this probably looks by sector:

Manufacturing

Not too much replacement from AI here except for some software maintenance and employee oversight. Once sufficiently advanced robots come around though, most jobs are gone for good (because the constraint even today is more efficacy than price).

This is a very positive thing. Almost all jobs that require repetitive human labor suck.

As this happens, prices drop overall & the demand for human-produced goods gradually increases. The DTC winners are increasingly producers that can warrant the highest brand premium.

Service

Prices drop dramatically for all unspecialized and modestly specialized skills, but over time. Highly specialized services still warrant significant human involvement from the best in the world.

The President’s personal lawyers are still mainly humans & heart surgeries in the richest neighborhoods of California are still performed by human surgeons.

As the price gap between decent AI-first firms and specialized human-first ones increases though, it is unclear to me exactly how much AI will win out.

Software

This category refers to any high-gross margin business with a build-once, deploy-infinitely model with negligible marginal cost/user. This includes data as a service, paid content, etc.

Honestly, the biggest problem with software seems to be the built-in switching cost. If you make the sale with a product good enough, most firms using your software are not incentivized to look for competing solutions - even when they start being built with AI and are far more affordable - until that’s exceedingly obvious or your product’s value/price stops being intrinsically justifiable.

This makes SaaS a really good business to be in, but regardless of whether switching software solutions becomes easier in the future - autonomous AI means more solutions in every space and generally lower prices across the board.

What This Means

Everyone gets richer!

In every one of these segments, prices for everything are lowered with automation. This means more firms come up as unspecialized offerings get commoditized and cheaper, and these new firms move into more 1. specialized roles or 2. management roles for markets with uncapped demand and luxury products (part of this is, again, going to be the elevation in the status of human-made goods, the same way we value nationally-made goods).

And as prices lower for virtually everything we need and want, everyone’s money goes further. Everyone gets richer.

How this fails: If it turns out to be that most people aren’t well equipped for management in the manufacturing sector OR if specialized service is where humans are most needed & enough humans fail to specialize into new roles quickly enough.

Naturally, now, the biggest problem is the short-term transition pain of adapting workers to an AI-based world. But I think we’re doing pretty well so far.

What Jobs Go Away?

While I think populist people will be replaced by people using AI narrative is broadly correct, there’s more to it.

It may then be natural to want to apply the cost/profit center segmentation, and while that’s a very useful guiding instinct there is a better way to differentiate here. Here’s how to know if your job is going away:

Demand Constrained: Jobs in sectors like accounting, customer support, and HR are demand-constrained. No firm really wants more people in these roles, just the bare minimum to continue operation. These are bad sectors to be in.

Non-Demand Constrained: Jobs in sectors like sales, R&D, and conditionally engineering, are not typically demand-constrained. If firms could have more of these roles for the same cost as before, they would generally take them. These are good sectors to be in.

The Obvious Nuance: Both camps of jobs here are not perfectly defined and do depend on the firm (some firms may decide to invest more in engineering while others pocket the difference). It’s just important to acknowledge the spectrum with an honest analysis of where different sectors lie on it.

Doomsday?

It’s possible. Everything I’ve said so far depends on AI not ending humanity. I also think current models are safe enough and mainstream AI teams seemingly responsible enough for me to be pro-acceleration today (but anti-open source).

If AI does end up eliminating humanity, I imagine it’ll be via human bad actors accessing incredibly powerful AI systems they can manipulate. Not via evil, autonomous, sentient AI directly.

The Conclusion

AI is great. As long as we’re extremely proactive with re-education, prepared to embrace fragmentation of firms and more entrepreneurship, and democratize safe access to the strongest model at any given time.