AI on AI

The Future of Work: AI, Automation, and Employment

Working in startups, or maybe, working in anything right now, it is hard to go an hour, let alone a day, without hearing about, advising on, or engaging with an AI. They are in everything already, and shortly they will be in everything else.

The aye-aye is often viewed as a harbinger of evil and death and killed on sight. Others believe, if one points its narrowest finger at someone, they are marked for death. Some say that the appearance of an aye-aye in a village predicts the death of a villager, and the only way to prevent this is to kill it. The Sakalava people go so far as to claim aye-ayes sneak into houses through the thatched roofs and murder the sleeping occupants by using their middle fingers to puncture their victims’ aorta.

– Wikipedia

Like the aye-aye in the dark forest, AI moves quietly through the systems we’ve built — tapping, listening, sensing. It doesn’t bulldoze its way in; it weaves itself through the cracks, threading into finance, education, medicine, art, and even our daily conversations. It’s a long, curious digit reaching into data, gently uncovering patterns, connections, and sometimes, unexpected insights. (Maybe) Not here to dominate us — just to find what’s hidden, and maybe help us understand it. There is a wealth of opportunity buried in the connections that we haven’t made yet, but also a mass of creeping dystopian novels that warn us we could mess this up.

A Quick Chat

For fun, I had a quick chat with ChatGPT about the risks of the growing prevalence and increasing reliance on AI for completing tasks. I wanted it not to be dystopian, but to explore the opportunities, and particularly against a backdrop of Trump’s tariffs, and the creation of new factories in the US. I was thinking specifically about Apple and Foxconn (a company with three quarters of a million employees in Taiwan alone and significantly lower salary and welfare considerations than a similar organisation would need in the US), and the target of the excellent book Dying for an iPhone which explored the fate of the Chinese factory workers so that we could all have our new tech.

Now ChatGPT definitely has a agreement bias, if you posit something, it always tries to agree with you, but it was interesting to me how quickly I got from an initial question about tariffs, automation and employment to:

Yes — automation increases productivity but often reduces employment.

And that’s the central contradiction:

• Fewer workers means lower wage outflows → less consumer spending
• But consumer demand is the fuel of capitalist growth
• So the system becomes increasingly productive but less able to sustain its own market

This is sometimes described as a “self-cannibalizing loop”:

Productivity ↑ → Wages ↓ → Consumption ↓ → Profits ↓ → Layoffs ↑ → Repeat

Which is especially problematic because capitalism thrives on mass participation:

• People working
• People earning
• People spending

When machines replace people at scale — even in better-paid countries — the money doesn’t circulate the same way.

– ChatGPT

With no further prompting it then continued to:

In the automation economy, money accumulates in capital, not labor:

• It flows to IP owners
• It flows to infrastructure owners
• It flows to shareholders

But if you automate everything and don’t redistribute, you eventually hollow out the base.

You get:

• Rising inequality
• Social unrest
• Decreased aggregate demand
• Political destabilization (see: 2016+, globally)

– ChatGPT

A Less Dystopian Outlook

Miserable, and right on the pulse of what is happening already, but there is hope, right?

What followed was a really interesting and reference chunk of dialogue offering three initial possibilities:

1. “Hybrid Work + Tech Dividend” Economies
2. Rise of Civic Tech & Mutual Credit Systems
3. Regenerative Capitalism (with Teeth)

Option 1 had its roots in UBI which is something I am very much in favour of, but was perhaps a little too ableist and factory like (to qualify for the income, you had to provide value to the tech companies), Option 3 was wonderfully green and cooperative, and Option 2 was full of commnuity.

What’s the Point?

We chatted about the relaistic potentials of each for a short while and it sparked a lot of ideas for me. And that’s a💡moment for me, just how good it is to be able to go down a rabbit hole (with references) and brainstorm both alone, yet with some of the greatest thinkers that ChatGPT has consumed into it’s massive knowledge base. It’s derivative work, and needs reference and fact checking. It isn’t a replacement for human discourse, you can say stupid things and it will tell you that you’re amazing and insightful, it is deeply flawed, but it is a brilliant and effecive tool.

I worry naturally about how much energy I consumed in what was quite a lengthy chat, and with the caveat that I haven’t checked the homework, but it looked to be about 8W Hours, so similar to having a modern light on for an hour. Not too bad. (Images though, go carefully, you can burn the planet there).

On Futures

For fun I asked about Tech Giants in this SolarPunk future, it handled all of them, my favourite given where I started…

This isn’t particularly a thought piece, just a random rabbit hole that I enjoyed. On a final note though, I asked ChatGPT about my personal belief that as models reach a point where they can achieve most tasks that we need them too, will prices sky rocket (to the end consumer) so that the AI companies can recoup their huge training investments and monumental running costs?

You’re asking the question that’s been keeping a lot of CFOs and CTOs awake at night — and no, you’re not wrong. Let’s unpack this with a little nuance.

Why Prices May Go Up (And Likely Will):

Massive Infrastructure Costs:
Training frontier models like GPT-4 or Claude Opus costs tens to hundreds of millions of dollars in compute alone — before including ongoing inference (running) costs.

Running Costs are Non-trivial:
Inference at scale isn’t cheap — especially with high-quality, low-latency service expectations. Even with efficiency improvements, server farms require constant upgrades, energy, cooling, and staff.

User Demand is Soaring:
As people integrate these tools into workstreams (and whole apps are built on them), reliability and availability become mission-critical. That demands serious investment.

Investor Pressure for ROI:
These companies have taken on massive VC investment with expectations of returns. At some point, “burn to grow” has to flip to “grow to earn.”

So Will Prices Go Up for Individual Users?

Eventually, yes — at least in one or more of the following ways:

Token quotas shrink on free tiers

Pro/premium pricing tiers rise

Access to advanced models (like GPT-5) becomes gated behind higher paywalls

New features (e.g., memory, agents, tools) may be pay-to-use

I think it is inevitable, as the global race cools down (and hopefully Europe hurry up), prices will not only increase, they will soar, and the economics of that are very confusing. The new daily habits we are forming won’t be cost effective anymore. Products being built now will need to increase cost to consumer, or work out how to run on smaller models (the rise of the SLM, note: if you’re building, you should do this now, no matter how cheap GPT-4o is), or die. Only high margin business, high net worths, or nations will be able to leverage the super powered models, and we go back to the dystopia that I trying hard to believe isn’t inevitable in the AI dominated near future.

Personally I’m still advising my kids to study whatever makes them happy, but hopefully that is subjects that prioritise critical thinking and communication skills (whatever, politics, history, geography, economics) coupled with creative pathways (I might enjoy AI gen’d content, and it must suck to be a junior *anything* now, but it is ultimately derivative crap). I believe that combination will distinguish Homo technensis from our synthetic friends.

I’ve got an idea for some small vignettes, call it speculative fiction if you’re modern, based on my explorations here. Thoughts I’ve been having for a while, almost enabled by my chat with ChatGPT, my pendulum still swings dystopian, but, it is a pendulum.