Sierra argues customer context is a proprietary moat, captured inside its Context Engine. This knowledge graph models that thesis alongside an agent-authored alternative: context as loosely-coupled Semantic Web infrastructure, where the differentiator is pluggable context quality, not platform lock-in.
KG curated by kg-generator + rdf-infographic-skill (Claude Sonnet 5) on behalf of Kingsley Uyi Idehen.
Sierra frames the competitive debate as a shift from "what can I do with AI?" to identifying where differentiation truly lies: intelligence is rentable, context is uniquely owned.
"Unlike intelligence, which every company can rent, this context is uniquely yours."
Sierra, Context Engine article
"How do companies avoid ceding value to a handful of models?"
— Satya Nadella, cited in the article
Sierra's Context Engine compounds customer context inside its own platform to power long-running Horizon agents. The alternative thesis argues the same compounding benefit is achievable without vendor lock-in, by loosely coupling five concerns as independent, standards-addressable layers.
Under a loosely-coupled Semantic Web architecture, the durable advantage is not who owns the biggest proprietary context store, but who has the best solutions for plugging into standards-based context infrastructure — identity, authentication, authorization, data spaces, and shared ontologies, addressable via open protocols, not a single vendor's platform.
Identity, authentication, authorization, and data spaces solve plumbing — getting data from A to B without a middleman. They don't solve meaning. Sierra's compounding loop discovers statistical, probabilistic context rules from accumulated interactions — useful, but not individually auditable and not guaranteed reproducible run to run. An ontology published as Linked Data (RDFS/OWL, validated with SHACL) instead supports deterministic reasoning: the same class hierarchy and equivalence axioms produce the same inferred fact every time, for any reasoner, regardless of vendor. The two forms of compounding are complementary — but only the ontology-backed form is portable across agents.
This reasoning is not a separate offline step. In a reasoning-capable engine, entailment is applied directly inside the query solution production pipeline — a single SPARQL SELECT/CONSTRUCT/DESCRIBE already returns inferred facts alongside asserted ones. And SPARQL isn't the only door in: the same reasoning-capable data space is also reachable via SPASQL (SPARQL patterns embedded in SQL, for SQL-native tooling) and GQL (ISO/IEC 39075, for graph-native tooling) when Virtuoso is the engine — one entailment-aware substrate, three standards-based query surfaces.
Sierra distinguishes business-known data from agent-learned data, and insists raw access is not context.
"Having access is not the same as having context."
Sierra, Context Engine article — device telemetry only becomes actionable once connected to order shipment, purchase history, and open support tickets.
Every agent decision is treated as an experiment. A small share deliberately test promising alternatives instead of defaulting to proven options, feeding a loop that discovers new context rules tied to business outcomes: saves, lifetime value, qualified leads.
"By morning, it knows a little more about what keeps your customers than it did the night before."
Sierra, Context Engine article
Sierra's central claim: foundational models cannot be a moat, since competitors access identical versions. The true moat is the accumulated record of observations and decision outcomes.
"That record is your moat, one your competitors cannot buy or shortcut."
Sierra, Context Engine article
The critical perspective does not dispute that compounding interaction history is valuable, or that raw access isn't context. It disputes that this value must be captured by locking context inside one vendor's application — a Semantic Web built on Linked Data principles can compound the same signal as a federated, portable record instead of a proprietary one.
This isn't a hypothetical architecture. Two systems already run this way — including the one that generated this very page.
One canonical identity, one preference set, one howto knowledge base — stored as plain RDF/Turtle files with an optional SPARQL endpoint over Virtuoso. No single vendor's agent owns it. The ontology layer isn't hypothetical either: a companion TBox alignment registry harmonizes 656 locally-minted classes/properties found across the wider corpus — 86 recurring in two or more independently authored documents — via owl:equivalentClass/owl:equivalentProperty, loadable into Virtuoso and exploitable by rule-based reasoners (Virtuoso inference rules, FuXi) for deterministic entailment across documents that were never designed to interoperate.
Runtimes observed plugging in:
Demonstrates: Identity, Identification, Data Spaces, and Ontology layers.
A single standards-based support skill — 25 MCP tools over SPARQL/SQL/GraphQL, v1.4.2 — resolves support cases by routing to whichever backend data space is selected at runtime, including a dedicated /troubleshoot workflow. No case data lives inside the skill itself. Its execute_spasql_query tool is direct evidence of query-language pluggability: the same reasoning-capable store is reachable via native SPARQL or via SPASQL-embedded-in-SQL, and Virtuoso maps the same substrate to GQL for graph-native tooling.
Data spaces it routes to at runtime: Demo, URIBurner (production), Localhost
Deployed unmodified on: Claude, Grok, Codex, Gemini/Antigravity, Vibe
Demonstrates: Authorization, Data Spaces, and Ontology layers.
Neither case study required Sierra-style vertical integration to compound value over time. Both separate what the context/capability is from which application is currently using it — the identity, the knowledge, and the data space all outlive any single agent session, runtime, or vendor.
Interactive graph of every entity and relationship in the companion RDF — Sierra's Context Engine components alongside the alternative infrastructure layers. Click any node or edge label to resolve its full description via URIBurner.
Fourteen questions covering both Sierra's argument and the alternative infrastructure thesis.
Terms from Sierra's article alongside the W3C/IETF standards underpinning the alternative thesis.
Five steps mapping each infrastructure layer to a concrete standards-based practice.
This page was generated from a Knowledge Graph transforming Sierra's Context Engine article into RDF using the kg-generator Thesis & Framework Article template, with an agent-authored critical-perspective section arguing for context as loosely-coupled Semantic Web infrastructure. The HTML infographic was rendered with rdf-infographic-skill, powered by Claude Sonnet 5. The knowledge graph is queryable live via the URIBurner SPARQL endpoint.
Technology Stack: