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

**Authors:** Raj Navakoti, Saideep Navakoti  
**arXiv:** [2603.14057](https://arxiv.org/abs/2603.14057)  
**Date:** 2026-03-14  
**Summary page:** https://www.summarizepaper.com/en/paper/2603.14057v1/

## Key points

- Large language model agents face challenges in performing enterprise-specific tasks due to a lack of domain knowledge
- Current approaches to addressing this issue: top-down knowledge engineering and bottom-up automation
- The Demand-Driven Context (DDC) methodology proposed by authors Raj Navakoti and Saideep Navakoti
- Inspired by Test-Driven Development
- Focuses on using agent failure as a signal for curating domain knowledge
- Presents real problems to agents, allowing them to demand the context they need to succeed
- Aims to build a sufficient knowledge base for specific domain roles within 20-30 cycles through minimal required knowledge curation
- Effectiveness of DDC demonstrated through a case study in retail order fulfillment with nine cycles resulting in a reusable knowledge base comprising 46 entities
- Proposed scaling architecture for enterprise adoption involving semi-automated curation and human governance

## Summary

In their paper titled "Demand-Driven Context: A Methodology for Building Enterprise Knowledge Bases Through Agent Failure," authors Raj Navakoti and Saideep Navakoti address the challenge faced by large language model agents in performing enterprise-specific tasks due to a lack of domain knowledge. This includes terminology, operational procedures, system interdependencies, and institutional decisions that are often considered tribal knowledge. The current approaches to addressing this issue fall into two categories: top-down knowledge engineering and bottom-up automation. Top-down knowledge engineering involves documenting domain knowledge before agents use it, leading to bloated and untested knowledge bases. On the other hand, bottom-up automation allows agents to learn from task experience but cannot acquire knowledge that exists solely in human minds. To overcome these limitations, the authors propose the Demand-Driven Context (DDC) methodology. Inspired by Test-Driven Development, DDC focuses on using agent failure as a signal for curating domain knowledge. Instead of pre-curating extensive knowledge bases, DDC presents real problems to agents, allowing them to demand the context they need to succeed. By curating only the minimum required knowledge through problem cycles, DDC aims to build a sufficient knowledge base for specific domain roles within 20-30 cycles. The authors demonstrate the effectiveness of DDC through a case study in retail order fulfillment where nine cycles targeting an SRE incident management agent resulted in a reusable knowledge base comprising 46 entities. Additionally, they propose a scaling architecture for enterprise adoption that involves semi-automated curation and human governance. Overall, the Demand-Driven Context methodology offers a novel approach to building enterprise-specific knowledge bases by leveraging agent failure as a guide for acquiring essential domain knowledge efficiently and effectively.