AI boosters on LinkedIn will have you belive AI will be taking everyone’s job tomorrow, but the anti-AI voices however still say never.
Full Show Notes: https://zexulo.xyz/2025/06/07/2513-gpt-job-losses-slowly-then-suddenly/
Experience.Computer: https://experience.computer/
Worldrunning.guide: https://worldrunning.guide/
Subscriber Zine support the show! https://startselectreset.com/Permanently moved is a personal podcast 301 seconds in length, written and recorded by @thejaymo
Subscribe to the Podcast: https://permanentlymoved.online/
Quarterly zine; my gift to you ✉️
GPT Job Losses: Slowly, Then Suddenly
There’s a hum of change in the air, AI is percolating through society.
Many are now asking not if AI will significantly impact employment, but when.
A few months ago, I was asked to contribute to a report on AI and technical unemployment. That report it seems, is now unlikely to ever be published. So, I thought I’d do something rare and stick my neck out and share my timeline for GPT-driven job losses.
When I say GPT, you likely think Generative Pre-trained Transformers, the LLMs making headlines. But to grasp the shape of what’s coming, we must consider another kind of GPT: General Purpose Technology. LLMs are both, joining a list that includes: the steam engine, common screw fittings, computers, and the Internet.
But the GPT whose history best models what’s coming is the Shipping Container.
This humble metal box offers us a blueprint for AI’s potential and eventual “slowly, then all at once” disruption.
For most of human history, loading and unloading boats was a labor-intensive ballet of muscle and coordination. Dock workers, Longshoremen, packing all the goods on by hand, then another crew taking them off again piece by piece at the other end.
I can’t cover the full history of the container, only its modern timeline is key. But the full story runs from Roman amphora, through coal boxes during the Industrial Revolution. The father of the modern shipping container however is Malcom McLean. A man with a post-WW2 vision for global intermodal transport, who launched the first modern container ship in 1956. It promised to slash costs and shipping times, but the revolution wasn’t immediate.
Google’s 2017 paper “Attention Is All You Need’ marks our own GPT revolution’s symbolic start — a date already fast fading into the past. For containers, their first eight years into the early 1960s were a “slowly” phase, mirroring AI’s current percolation. Efficiencies were clear, but adoption was sluggish due to the need for massive infrastructural investment. New cranes, redesigned ports, specialised ships, and crucially, a lack of universal standards.
Longshore unions, already seeing the change unfolding in slow motion, negotiated agreements like the 1960 Mechanisation and Modernisation Agreement, trading benefits for accepting the new technology. Any job losses at the time were masked by attrition and retirement.
In 1955, New York and New Jersey ports employed around 35-40k longshoremen. By 1965, 9 years after the containers debut, that number was still around 30-35k. The true scale of the impending disruption was still incubating.
Then came the “all at once” phase, triggered by Standardisation. Between 1968 and 1970, the International Organization for Standardization (ISO) established common standards for container sizes, corner fittings, and handling mechanisms catalysing critical mass adoption and network effects.
Interoperability transformed a collection of previously proprietary systems into a cohesive, global network and helped create the modern world. The more entities adopted the standard, the more valuable the system became for everyone else, compelling them to join. Investment in infrastructure exploded because there was now a universal, predictable system to build upon.
This all suggests LLMs could begin reach societal maturity around 2029, twelve years on from 2017. What constitutes ‘standardisation’ for AI of course remains to be seen. Maybe it’s common safety protocols, or APIs that allow AI to plug into existing corporate systems with ease. Many of which are emerging now.
Once converged, AI will shift from fragmented tools to an integrated, trusted, easily deployable ecosystem. Network effects take hold, and the business case for automation becomes overwhelming.
Considering typical large IT project timelines and infrastructure build-outs (data centres, power etc), 2029 for AI’s maturity seems reasonable. Just 2 years after AGI is apparently going to arrive.
Back to the docks: In 1970, 20k people still worked in the New York port system. But just five short years later, only 3,500 longshoremen remained. A 90% drop in the “all at once” phase. If our timeline holds, this suggests a similar period of mass unemployment for knowledge workers around 2036.
AI boosters on LinkedIn will have you belive everyone’s job is vanishing tomorrow, anti-AI voices say never. But AI researchers, when pressed, often ballpark a timeframe about 10 years from now. Which aligns with our containerisation thesis.
Of course, AI is software, not physical infrastructure; the timeline might compress; but not by much, considering human and organisational inertia. Think too about the workplace changes of the 90s: the introduction of the desktop computer and networking. Typists, internal mailrooms, and secretarial pools saw tens of thousands of jobs evaporate within a decade.
I personally view AI-driven unemployment as an imminent concern. A possibility to take seriously. It’s only two parliamentary terms away.
Recently, on a Zoom call, I heard an AI call centre software founder boldly state that every millisecond shaved from agent response times was closer to ‘good enough’ for an executive to pull the trigger on adoption. Each millisecond, he claimed, was roughly 100,000 jobs lost worldwide.
The hum of change is palpable, but not yet a roar. The critical question remains: what will AI’s “standardisation” moment be? the point that the percolation becomes the flood? Whatever the catalyst, ’the neoliberal bullshit jobs’ that have propped up the economy since 2008, are in the crosshairs.
Containerisation’s slow creep lulled many into false security, underestimating the eventual velocity and magnitude of job displacement. We must be wary of similar complacency regarding LLMs.
The next question is: what will all these people do next? Will our governments have us fight a stupid war? or climate change?
After all, robots and AI can’t yet plant forests, restore ecosystems, or tend gardens.
Subscribing to SSRZ supports my online work and creative projects.
As a thank you, I send you my zine four times a year, just like it’s 1994.
No spam. No email. Cancel at any time.
Leave a Reply