Theory background
Transaction cost economics asks why some activities are handled through markets while others are organized inside firms. The classic answer is that using markets is not frictionless: search, negotiation, monitoring, coordination, and enforcement all create costs.
For managers, the practical point is that a task should not be judged only by the price of a tool or vendor. It should be judged by the total cost of coordinating the work reliably.
Translate to AI adoption
AI changes the boundary between buying, building, and delegating work to software agents. A company may outsource an AI project, build an internal workflow, or use an agent to perform repeated tasks under supervision.
Each choice carries different transaction costs. Outsourcing can reduce technical setup but increase communication and verification costs. Internal development can preserve context but require scarce skills. Agentic workflows can reduce repeated handoffs but increase monitoring and governance requirements.
Managerial implication
The make-or-buy question for AI should be framed as a boundary decision. If the workflow is highly specific, changes often, touches sensitive data, or depends on tacit internal judgment, full outsourcing may become expensive even when the vendor price looks attractive.
If the workflow is standardized, separable, and easy to evaluate, external tools or managed services may be more efficient.
What to test
Before scaling, estimate the hidden coordination costs: how many clarifications are needed, how often outputs require review, who can verify quality, and whether source data can be shared safely.
A useful AI trial does not only test whether the model can perform the task. It tests where the firm boundary should sit after AI is introduced.