The Knowledge Horizon: Accountability, Expertise Erosion, and the Case for Human Orchestration in Agentic AI

As agentic AI systems move from monolithic frontier models to specialised orchestrated architectures, the primary obstacle to full automation is no longer capability alone, but accountability. Current consequential institutions require a human-accountable actor in the chain: someone who can bear responsibility, hold a licence, sign off on work, and answer for failures. Human orchestration therefore appears necessary not merely as a temporary safeguard, but as the most institutionally viable bridge currently available between agentic efficiency and institutional legitimacy. The paper’s contribution is threefold. First, it argues the accountability case through institutional mechanics rather than moral philosophy, identifying the accountability gap as a compound deficit with two separable components (one architectural, one environmental) that may require different interventions. Second, it proposes adversarial orchestration as an architecture that satisfies the accountability constraint while preserving cognitive engagement. An accidental benefit emerges: the accountability requirement, by keeping humans in the loop, inadvertently slows the expertise erosion that full automation would accelerate. Third, it documents the expertise pipeline problem created by skill encoding: when routine tasks are automated, the developmental pathway through which junior professionals build the judgment needed for senior roles disappears. Once that bridge is acknowledged, a second and longer-horizon question emerges: whether keeping humans in the orchestration loop also serves to preserve the expertise pipeline that large-scale skill encoding would otherwise erode. The scope of orchestration judgment extends beyond model output to the distribution infrastructure through which AI tools reach users. A recurring pattern of infrastructure failures in AI tool distribution (documented in detail in Paper 2) illustrates why the orchestrator must exercise judgment on the entire pipeline, understand the structural reasoning behind that judgment, and be able to transmit that reasoning to a successor. Paper 3 of 5 in the Confidence Curriculum series 10.5281/zenodo.19226032.