I (Rizwanul Islam Afraim) recently published a paper on SSRN titled Agentic AI as Coordination Infrastructure Technology: Structural Implications for Firms, Growth, and Economic Divergence.
The paper came out of a growing discomfort I had with how AI is being discussed in public.
A lot of the conversation is not wrong. It is just shallow.
We keep circling around familiar questions:
Will AI improve productivity?
Will it automate tasks?
Will it replace jobs?
Will it help companies do more with fewer people?
Those are valid questions. But they still sit too close to the surface.
What I think is being underestimated is the degree to which agentic AI changes coordination itself.
That is a deeper layer than task automation.
Modern firms and modern economies are not constrained only by production. They are constrained by handoffs, supervision, information routing, managerial bandwidth, compliance friction, and the countless forms of coordination that sit between intention and execution.
That is why firms need layers. That is why systems get heavy. That is why scale often creates drag.
If AI begins to reduce those coordination costs directly, then we are not just looking at a smarter software layer. We are looking at a possible restructuring force.
That is the motivation behind the idea I introduce in the paper: Coordination Infrastructure Technology (CIT).
The point of the term is to move the discussion away from “AI as a tool that helps people complete tasks” toward “AI as infrastructure that changes how workflows are coordinated.”
That shift has consequences.
It suggests that the biggest transformation may not be who writes faster, summarizes better, or drafts quicker. It may be who can redesign firms, systems, and institutions around compressed coordination cost.
That is also why I develop the Coordination Compression Hypothesis (CCH). If agentic systems can increasingly handle workflow routing, tracking, supervision support, documentation, and other coordination-heavy functions, then the number of optimal coordination layers inside organizations may decline.
And once that begins, firm structure itself becomes part of the AI story.
Then comes the labor side.
The effects of AI will not be distributed evenly. That is why I introduce the Dynamic Agentic Productivity Gradient (DAPG). Some people will use AI casually. Some will build workflows around it. A smaller group will orchestrate entire systems with it. The distance between those groups matters. It affects income, leverage, and long-run advantage.
I also think this matters at the national level more than many policy discussions currently admit.
Countries will not benefit equally from the AI transition just because AI tools become globally available. The deeper question is whether they become AI-orchestrating economies or remain AI-consuming economies.
That distinction may shape:
- where digital rents flow
- who accumulates intangible capital
- who owns the infrastructure layer
- who becomes dependent on foreign systems
For countries like Bangladesh, that question is not abstract. It is strategic.
There is real opportunity in localized AI services, orchestration capability, SME transformation, workflow design, and regional value creation. But there is also serious risk in staying concentrated in routine, lower-margin digital work while building on infrastructure that others own.
That is the broader concern running through the paper.
My goal was not to exaggerate AI. It was to frame it more accurately.
If we keep treating agentic AI only as a productivity enhancement story, we risk missing the deeper reordering underneath: coordination, hierarchy, dependency, and economic divergence.
That is the line of thought I wanted to put into public form.
If you want to read the full paper, here it is:
https://ssrn.com/abstract=6236898