The speed with which enterprise agent infrastructure has constituted itself as a distinct venture capital category is, even by the standards of the current AI investment cycle, striking. Twelve months ago, most investment theses in this area were positioned as adjacencies — AI safety tooling, developer infrastructure, enterprise workflow automation. Today, according to an analysis of PitchBook data aggregated by Terekhin Digital Media, the category has absorbed more than four billion dollars on a trailing twelve-month basis, and the firms deploying that capital have, by and large, converged on a coherent view of where value will accrete.

The concentration of capital across subcategories is uneven in ways that reveal investor conviction about the value stack. Agent orchestration frameworks — the infrastructure layer that coordinates multi-agent workflows, manages context, and handles the handoffs between AI systems — have attracted the largest allocations, with a median round size of sixty-two million dollars and several transactions exceeding two hundred million. The implicit thesis is that orchestration will become the operating system of enterprise AI: the layer through which all other components interact, and therefore the layer with the greatest potential for switching costs.

Evaluation and observability tooling has attracted the second-largest pool of capital, which reflects a recognition that the enterprises deploying agents need infrastructure to understand what those agents are actually doing. The analogy to application performance monitoring in the SaaS era is frequently invoked by investors in this segment: APM became indispensable infrastructure for software operations, and the firms that provided it — Datadog, New Relic, Dynatrace — generated enormous value over time. Whether the analogy holds depends on whether AI observability proves to be a distinct layer or becomes absorbed into existing monitoring platforms.

The category that has attracted the least capital relative to its apparent importance is agent safety and governance tooling — the infrastructure for defining what agents are and are not permitted to do, auditing their actions, and providing human oversight mechanisms for high-stakes decisions. Investors cite the difficulty of monetising governance tooling as the primary constraint: enterprises want safety, but they are not consistently willing to pay separately for it when it is expected to be embedded in the orchestration layer.