A few weeks ago, IBM had its worst single day on the stock market in 25 years. The trigger was not an earnings miss or a product failure — it was a blog post from Anthropic. Not even a real announcement: a speculative piece framed as a hypothetical 2028 internal memo about AI writing COBOL replacement code. Markets read it as an existential threat to IBM’s mainframe business and sold hard.
That moment is clarifying. Not because it told me anything about IBM’s future — it told me something about how thin the ground is under most AI economic analysis right now. We are pricing narratives, not fundamentals. And the narratives are incoherent.
Wall Street has somehow managed to price two contradictory bets at once: AI is too weak to justify the infrastructure spending going into it, and too strong for any existing software category to survive. That kind of incoherence usually means the underlying model is wrong, not just uncertain.
The doom scenario is not wrong on technology
The formalized version of the doom argument runs like this: AI displaces knowledge workers faster than new roles appear, reducing aggregate purchasing power and triggering demand contraction. It is internally consistent. The historical record of technological displacement has always shown more jobs created than destroyed — but always on generational timelines, not quarterly ones. The doom scenario requires displacement to happen faster than that.
Here is the part worth sitting with: the economist who built this model out most rigorously concluded that its own required conditions are probably too extreme to hold. Not because the technology is not capable enough. Because societal inertia will not allow it to deploy that fast.
The scenario is wrong on economics, not on technology. That distinction matters more than it might seem.
The force both narratives miss
There are two curves worth putting on the same chart. One is AI capability: reasoning depth, the ability to sustain complex tasks over time, cost efficiency. That curve is steep and accelerating.
The other is organizational dissipation — the rate at which organizations actually reorganize their workflows, processes, and roles around new AI capabilities. That curve is much slower. And it is constrained by friction forces that neither the doom nor the boom scenario takes seriously enough.
The gap between those two curves is where the real story is.
Why dissipation lags so far behind capability
Regulatory inertia is the most underestimated force. Financial services need regulatory approval. Healthcare navigates compliance frameworks — privacy law, device clearance, institutional review — that run on timelines measured in years. COBOL systems process the majority of financial transactions today. The bottleneck to replacing them is not whether AI can write the replacement code. It is whether the institutions responsible for those systems can absorb the change.
Organizational inertia runs deeper than most technology optimists acknowledge. Workflows and roles are designed around current capabilities. Redesigning them requires change management, retraining, and process reengineering. None of that compresses because the models improved.
Institutional trust has to be built incrementally. Deploying AI at scale with appropriate oversight and audit trails takes time. A strong benchmark score does not purchase trust.
Integration complexity is the chain from “technically possible” to “the economy has reorganized around this” — and there is friction at every link.
Where professional value actually concentrates
If both the doom and boom scenarios underestimate dissipation, then the gap itself is where value concentrates during the transition. The people who can take what AI can technically do and make it actually usable inside real organizations — who understand the friction forces well enough to work around them — are in a structurally strong position for as long as that gap persists.
That framing is more durable than either narrative. It does not require you to know whether AI will ultimately be transformative or destructive. It just requires you to understand where you personally sit on the dissipation curve: which of your skills are in the zone where AI capability has already outpaced organizational adoption, and which are still in the zone where human judgment bridges the gap.
There is also unusual value right now in being able to explain this calmly to people who are panicking. The IBM story is going to keep repeating. The professionals who can walk into a room with a clear, evidence-based model of what is actually happening — and what would actually have to be true for the catastrophic scenarios to hold — are offering something genuinely scarce.
What happens next
The transition will be slower, messier, and more unevenly distributed than either scenario predicts. That is not optimism or pessimism. It is what the friction forces imply.
The most useful question is not “will AI take this job” but “where in the dissipation curve does this work sit, and what does that gap actually look like.” That is a question with an answer — and it is answerable without waiting to see which narrative eventually wins.
Text summarized and optimized using Anthropic’s models and reviewed by a human.