Sharpening the Axe That’s About to Cut You: Why Individual AI Survival Advice Isn’t Enough
Terry Sweetser | IEISI | March 2026
Shane Collins recently distilled the viral Citrini Research “2028 Global Intelligence Crisis” thought experiment into ten market realities and a three-step survival process (Collins, 2026). His advice — audit your income stream, upgrade your AI toolkit, push the limits of delegation — is practical, well-intentioned, and almost certainly what most readers need to hear right now.
It is also dangerously incomplete.
Not because the steps are wrong individually. They’re not. But because they treat a systemic, structural crisis as an individual career management problem. And the history of socio-technical failure tells us that individual optimisation without systemic redesign doesn’t produce survivors — it produces a more efficiently hollowed-out system.
The Composition Fallacy
Collins’ Step 1 — audit your income stream, pivot to roles requiring empathy, physical interaction, or strategic leadership — is individually rational. If you pivot, you improve your odds.
But what happens when everyone receives the same advice simultaneously?
Van Geelen and Shah (2026) describe exactly this dynamic in their original thought experiment: displaced senior product managers driving for Uber, compressing wages for existing gig workers, who were already struggling. When a Salesforce PM takes a $45,000 rideshare job, that doesn’t just affect her — it depresses earnings for every existing driver in that market.
This is the composition fallacy: what works for one person fails when everyone does it. If every threatened knowledge worker pivots to “empathy and strategic leadership” at the same time, those sectors flood with overqualified talent, wages collapse, and the safe harbour turns out to be another exposed coastline.
The advice isn’t wrong. It’s just not a solution. It’s triage.
Sharpening the Axe
Collins’ Step 2 — pay for premium AI tools, stop competing with outdated models — is the most revealing recommendation. Think about what it’s actually asking you to do.
You pay $20 a month. You become dramatically more productive. Your output quality rises. Your employer notices. They realise they need fewer of you. The productivity gains accrue to capital — to the shareholders, the platform owners, the compute providers — not to you.
You have just sharpened the axe that is about to cut you.
This isn’t a new dynamic. Socio-Technical Systems (STS) theory, developed by Trist and Bamforth (1951) studying British coal mine mechanisation, established the principle seventy years ago: when workers are asked to optimise the technical system without any agency over the social system that determines how the gains are distributed, the result is not shared prosperity. It is extraction followed by displacement.
Collins’ advice is to optimise the technical system. At no point does he ask who controls the social system — who decides how the productivity gains are shared, who sets the terms of employment, who funds the safety net when the displacement hits.
Understanding the Loom
Collins’ Step 3 — push the limits of delegation, use AI until you hit your daily limits — rests on an implicit assumption: that understanding AI makes you safe from AI.
Understanding the power loom didn’t save the handloom weavers. Understanding the longwall mining machine didn’t save the small coal teams. Understanding the automated teller machine didn’t save the bank tellers (though the ATM story is more complex than people think — it shifted their role rather than eliminating it).
Tool literacy is necessary. It is not sufficient. Knowing how to prompt Claude doesn’t change the ownership structure of the compute you’re prompting on. It doesn’t change the tax code. It doesn’t change the social contract. It doesn’t give you a vote on whether your employer replaces your team with an agent workflow next quarter.
Collins confuses individual capability with structural power. In STS terms, he’s advising people to become more effective components of the technical system — without asking whether the social system will keep them in the loop at all.
What Collins Missed: The System, Not Just the Self
Here’s the core problem. Every piece of AI survival advice currently circulating — Collins’, and the dozens like it — operates at the level of the individual worker navigating a changing market. Upskill. Reskill. Learn to prompt. Become a “centaur.” Find your “human moat.”
None of it addresses the structural question: who is redesigning the social system to match the technical system that’s being deployed?
Trist and Bamforth’s (1951) foundational insight wasn’t “teach the miners to use the longwall machine better.” It was that the technical system and the social system must be designed together — jointly optimised — with the people at the interface having genuine agency over how that interface is managed (Cherns, 1976). When engineers optimised the technical system alone, productivity dropped, absenteeism soared, and industrial conflict intensified. The technically superior system produced worse outcomes because it destroyed the social system it depended on.
When you have humans in the system, you have a human system.
AI deployment in 2026 is longwall mining at civilisational scale. The technical system is being optimised at extraordinary speed by engineers and capital markets. The social system — fiscal structures, welfare safety nets, democratic institutions, community networks, the entire web of human coordination — is being treated as an externality. As something that will “adapt.”
The coal mines tell us it won’t. Not without design. Not without agency. Not without the people at the interface having a genuine say.
A Different Three Steps
So here’s an alternative framework. Not instead of Collins’ advice — his steps are fine as personal triage — but in addition to it, addressing the systemic layer that individual action cannot reach.
Step 1: Know your system, not just your skills.
Don’t just audit your income stream. Map the system your income depends on. Who pays you? Where does their revenue come from? What happens to that revenue if their customers are displaced? What happens to your government’s tax base if your entire sector contracts? If you’re in Tonga, what happens to your family’s income if your cousin in Western Sydney loses their warehouse job to a robot?
The point isn’t to terrify yourself. It’s to understand that your exposure may not be where you think it is. A Fijian hotel manager’s risk isn’t that AI replaces them — it’s that AI displaces the Australian middle class whose discretionary spending funds their tourism season. A PNG community health worker’s risk isn’t automation — it’s that Australia’s fiscal stress triggers aid cuts. Understanding the system tells you where the actual vulnerability lies.
Step 2: Build collective capacity, not just individual capability.
Collins says upgrade your personal toolkit. Fine. But also ask: what is my professional association doing about the structural shift? What is my union, my industry body, my community organisation, my local government doing? If the answer is nothing — which it probably is — then the most valuable thing you can do is start that conversation.
STS theory is clear on this: joint optimisation requires collective agency. Individual workers optimising individually is not joint optimisation. It’s parallel isolation. The miners who successfully adapted to mechanisation did so as self-managing teams who negotiated the interface between the technical and social systems collectively (Trist et al., 1963). The miners who were reorganised into isolated, specialised roles by engineers who never consulted them saw their performance collapse.
The AI equivalent of the self-managing team is not a lone worker with a ChatGPT subscription. It is organised groups of workers, communities, and citizens who have a genuine seat at the table where the deployment decisions are made.
Step 3: Demand systemic redesign, not just personal adaptation.
Collins’ framework ends at the individual. But the crisis is structural. The tax base is eroding (Korinek & Lockwood, 2026). The development model is loading vulnerable countries onto the displacement curve (UNDP, 2025). The institutional architecture that could manage a global transition doesn’t exist yet.
Individual upskilling doesn’t fix any of that. What does? Political engagement. Institutional design. Demanding that AI deployment be jointly optimised with the social systems it transforms — which means demanding that governments restructure taxation, that development institutions audit their AI vulnerability, that employers negotiate transition plans with workers rather than presenting them with redundancy notices, and that the people most affected by AI displacement have a genuine voice in how it’s managed.
This isn’t idealism. It’s engineering. You wouldn’t deploy a new network architecture without testing it against the systems it connects to. You wouldn’t install a new power grid without coordinating with the water, transport, and communications systems that depend on it. Deploying AI into a society without redesigning the fiscal, welfare, and democratic systems that depend on human cognitive labour is the same category of error.
It’s just bigger.
Triage and Treatment
Collins’ advice is triage. It will help some people survive the next 18 months. It deserves to be taken seriously.
But triage without treatment just produces a longer queue at the emergency department. The treatment for a socio-technical crisis is socio-technical redesign — joint optimisation of the technical and social systems, with the agency of the people at the interface built into the design.
We know how to do this. The theory is seventy years old. The evidence is overwhelming. The principle is simple:
When you have humans in the system, you have a human system. Design accordingly.
References
Cherns, A. (1976). The principles of sociotechnical design. Human Relations, 29(8), 783–792. https://doi.org/10.1177/001872677602900806
Collins, S. (2026, February 25). The “2028 Intelligence Crisis”: Why a single Substack post just crashed the stock market. Activated Thinker (Medium). https://medium.com/activated-thinker/the-2028-intelligence-crisis-why-a-single-substack-post-just-crashed-the-stock-market-239e523c127d
Korinek, A., & Lockwood, L. M. (2026). The future of tax policy: A public finance framework for the age of AI. Brookings Institution / NBER Working Paper No. 34873. https://www.brookings.edu/articles/future-tax-policy-a-public-finance-framework-for-the-age-of-ai/
Trist, E. L., & Bamforth, K. W. (1951). Some social and psychological consequences of the longwall method of coal-getting. Human Relations, 4(1), 3–38. https://doi.org/10.1177/001872675100400101
Trist, E. L., Higgin, G. W., Murray, H., & Pollock, A. B. (1963). Organizational choice: Capabilities of groups at the coal face under changing technologies. Tavistock Publications. https://books.google.com.au/books?id=MxynAAAAIAAJ
UNDP Asia-Pacific. (2025). The next great divergence: Why AI may widen inequality between countries. United Nations Development Programme. https://www.undp.org/asia-pacific/publications/next-great-divergence
Van Geelen, J., & Shah, A. (2026, February 22). The 2028 global intelligence crisis. Citrini Research. https://www.citriniresearch.com/p/2028gic
Terry Sweetser is the founder of IEISI and serves as Chair of the APNIC Routing Security SIG and Secretary of PACIXP. He writes at the intersection of technical systems thinking and organisational leadership. His longer article, “The Largest Satisfice in Human History: AI, Socio-Technical Failure, and the Coming Institutional Crisis,” is forthcoming.
The main medium article is now up at https://medium.com/@terrysweetser_90287/the-largest-satisfice-in-human-history-ai-socio-technical-failure-and-the-coming-institutional-280529e603de