The Problem of Opaque Knowledge

Modern AI agents often struggle to interact with external knowledge bases because these systems lack standardized interfaces for 'discoverability.' When an agent needs to retrieve information, it often relies on imprecise semantic search or hard-coded API calls. This paper introduces a formal framework for 'Agentic KG Affordances,' which treats knowledge graphs not just as static data stores, but as interactive environments where agents can programmatically query the structure, schema, and capabilities of the data.

Defining Agentic KG Affordances

The core of the framework is the concept of an 'affordance'—a set of properties that define what an agent can do with a specific knowledge graph. By formalizing these affordances, the authors provide a mechanism for agents to:

• Self-Discover Schema: Instead of requiring a pre-defined prompt describing the database, the agent can query the KG to understand its own structure, relationships, and constraints.
• Negotiate Capabilities: The framework allows the KG to expose what types of reasoning or retrieval it supports, allowing the agent to adapt its strategy based on the available data environment.
• Standardize Interaction: By creating a formal protocol for these interactions, the framework reduces the hallucination risk associated with agents guessing how to query complex, proprietary data structures.

Implications for Agentic Architecture

This approach shifts the burden of knowledge integration from the prompt engineer to the system architect. Rather than building massive, monolithic RAG pipelines, developers can expose structured 'affordances' that allow agents to navigate data dynamically. This is particularly useful for complex enterprise environments where data is siloed and schema-heavy, as it allows agents to 'browse' the data landscape before executing specific retrieval tasks, leading to more reliable and context-aware outputs.