AI Lowers the Cost of Coordination, So Why Are Organizations Still Siloed?

I kept noticing the same pattern across the companies I was reading about and the enterprise AI tools I was tracking. Salesforce Einstein, Microsoft Copilot, Google Vertex AI, all of them marketed as tools that would connect teams, automate coordination, break down silos. And every implementation I looked at had the same structure. The AI was layered on top of the existing org chart. It helped people inside their existing departments do their existing jobs a little faster. It did not change who reports to whom, how budgets are allocated, or where decisions get made.

This bothered me because I had been reading a lot of transaction cost economics for my comps work. Coase (1937) asked why firms exist at all. His answer was that markets are not free. Searching for a supplier, negotiating a contract, enforcing an agreement, all of these carry costs that internal organization can reduce. Williamson (1975) turned this into a full theory of the firm. Organizations choose hierarchies over markets when transaction costs are high, especially when assets are specific to a relationship, when the future is uncertain, and when transactions repeat frequently enough to make governance design worthwhile. The whole reason we have managers, departments, approval chains, and silos is that these structures reduce the cost of coordinating work better than the open market can.

Gurbaxani and Whang (1991) applied this logic to information technology directly. They argued that IT reduces both internal coordination costs and external transaction costs, and the direction of the boundary shift, whether firms centralize or outsource, depends on which cost falls more. This is the part that matters for AI. If AI reduces coordination costs significantly, TCE predicts that organizations should either flatten, because the hierarchy becomes less necessary when coordination is cheap, or expand, because coordinating with external partners also gets cheaper. Either way, the organizational structure should change.

But here we are in 2026, and the dominant pattern is not restructuring. It is augmentation inside existing structures. Microsoft Copilot drafts emails, summarizes meetings, and generates documents for people who already have those responsibilities. It does not rewire who talks to whom. Salesforce Einstein suggests next-best actions for sales reps inside their existing territories and reporting lines. It does not merge the sales and marketing departments because the AI can now handle the coordination between them. The tools reduce friction at the edges without changing the organizational logic at the center.

I think there is a real theoretical distinction here that the TCE lens makes visible. Most enterprise AI tools are designed to lower coordination costs inside the firm. They help the marketing team coordinate with itself faster, not coordinate with sales in a fundamentally different way. TCE predicts that lowering internal coordination costs should make hierarchies more efficient, not less necessary. If the cost of managing people internally drops, the efficient boundary can actually shift toward hierarchy, not away from it. The Gurbaxani and Whang (1991) logic is symmetrical. When AI lowers external coordination costs more, it should push firms toward markets, outsourcing, and platform-based structures. When it lowers internal coordination costs more, it should make existing hierarchies work better without changing the structure.

The reason organizations are not reorganizing around AI may be that the AI tools being deployed today reduce internal coordination costs more than external ones, and the TCE prediction for that scenario is exactly what we see. Firms get more efficient hierarchies, not flatter ones. Copilot makes your department run faster. It does not make your department disappear.

But I think there is a second force at work, and it is harder to see from inside the TCE frame alone. The organizations that deploy AI do not face a clean make-or-buy choice for every coordination problem. They face legacy structures, existing power distributions, and professional identities tied to the current hierarchy. The head of marketing has incentives to keep the marketing department intact, even if the AI could handle half the coordination that the department exists to manage. Baird and Maruping (2021) argued that agentic systems shift the relationship from use to delegation, and delegation requires appraisal, distribution, and coordination. Organizations are not set up to delegate authority to AI systems because delegation is not just a technical decision. It is a political one. Who gets to decide what the AI coordinates? Who loses budget and headcount when a coordination function is automated? These questions are not in the TCE model, but they determine whether the structural change that TCE predicts actually happens.

If I were advising a company that is serious about getting value from AI, I would start with the TCE question before the technology question. Which coordination costs in your organization are highest? Between which teams? If you map the transaction costs of your internal workflows, the places where information gets dropped, where handoffs require three meetings, where approvals take two weeks, those are the places where AI can change the structure, not just speed up the existing process. But you would need to be willing to let the coordination pattern change, not just automate the current friction. If the AI makes it cheap for the design team to coordinate directly with the production team instead of going through a product manager, the product manager role should change. Most organizations are not willing to make that move.

My reading of TCE says that AI might be the largest reduction in coordination costs since the telegraph. If the theory is right, and I think it is within its boundary conditions, the organizational effects should be visible in the next few years. Firms that treat AI as a bolt-on productivity tool inside existing structures will get incremental efficiency. Firms that treat AI as a governance change, as something that rewires who coordinates with whom and how, are the ones that will look structurally different from their competitors. The value is not in the AI model. It is in the reorganization that the model enables. And most organizations are not doing that part yet.