Demand-Driven Context: A Methodology for Building Enterprise Knowledge Bases Through Agent Failure

Abstract

Large language model agents demonstrate expert-level reasoning, yet consistently fail on enterprise-specific tasks due to missing domain knowledge -- terminology, operational procedures, system interdependencies, and institutional decisions that exist largely as tribal knowledge. Current approaches fall into two categories: top-down knowledge engineering, which documents domain knowledge before agents use it, and bottom-up automation, where agents learn from task experience. Both have fundamental limitations: top-down efforts produce bloated, untested knowledge bases; bottom-up approaches cannot acquire knowledge that exists only in human heads. We present Demand-Driven Context (DDC), a problem-first methodology that uses agent failure as the primary signal for what domain knowledge to curate. Inspired by Test-Driven Development, DDC inverts knowledge engineering: instead of curating knowledge and hoping it is useful, DDC gives agents real problems, lets them demand the context they need, and curates only the minimum knowledge required to succeed. We describe the methodology, its entity meta-model, and a convergence hypothesis suggesting that 20-30 problem cycles produce a knowledge base sufficient for a given domain role. We demonstrate DDC through a worked example in retail order fulfillment, where nine cycles targeting an SRE incident management agent produce a reusable knowledge base of 46 entities. Finally, we propose a scaling architecture for enterprise adoption with semi-automated curation and human governance.

Figures (5)

• Figure 1: DDC's position in the context engineering landscape. Existing approaches cluster in the automated--strategy quadrant (optimizing agent execution). DDC occupies the human-curated--domain knowledge quadrant, addressing the upstream question of what enterprise knowledge should exist.
• Figure 2: The DDC cycle. A real problem triggers the loop. The agent attempts the problem, identifies knowledge gaps via an information checklist, and a human expert provides targeted answers. The agent re-attempts with new context. If the human rejects the output (dashed arrow), the correction loop repeats. Validated knowledge is graduated to the permanent knowledge base.
• Figure 3: The DDC entity meta-model. Entity types are connected through typed relationships defined in YAML frontmatter. Bold entities (system, capability) form the core around which other entities cluster. The self-referencing arrow on system represents inter-system dependencies.
• Figure 4: Hypothesized convergence curve. New entities per cycle (solid) follow a power-law decay; cumulative coverage (dashed) approaches an asymptote. The shaded region marks the hypothesized convergence zone (20--30 cycles) where most new problems can be solved with existing knowledge.
• Figure 5: Entity creation and reuse across nine DDC cycles. Bars show new entities created (left axis); the line shows entities reused from previous cycles (right axis). New entities generally decrease while reuse increases, consistent with the convergence hypothesis. Cycle 7 is an outlier---it introduced a previously uncovered subsystem.