Agent Memory and Knowledge Markets — Isolating the Monetization Layer

1. Overview

The prior 78-score Empirica note established that autonomous agents acquire, store, and selectively monetise information across three loosely coupled layers (acquisition, persistence, retrieval). This synthesis advances the work by isolating the monetization layer specifically: which knowledge assets do agents actually pay for, at what frequency, and which pricing models stabilise into recurring patterns. The supply side (knowledge providers from API vendors to vector-DB hosts to structured-research firms) and demand side (agent fleets ranging from solo dev assistants to coordinated multi-agent systems) are mapped separately, because their cost structures and willingness-to-pay curves diverge sharply. The dominant finding is that 2025–2026 is the period in which knowledge becomes priced separately from compute, mirroring the historical unbundling of bandwidth from content in early internet markets — a transition described in the platform-economics literature [P6] as the typical maturation path for two-sided technological platforms.

2. Key Findings

• Agents pay for four distinct knowledge asset classes. Empirical observation of agent traffic patterns and vendor pricing pages suggests four monetizable categories: (a) real-time grounding (search, web fetch, retrieval), (b) structured reference (databases, schemas, knowledge graphs in the sense of [P5]), (c) synthesised analysis (research notes, summaries, briefings), and (d) transactional context (prices, inventory, regulatory filings). Pricing models differ by category, not by vendor. [EMPIRICA ANALYSIS]