The Anatomy of the Agentic Namespace

Most analyses of AI-era domain names are anecdote — a headline sale, a hot TLD. This is the other thing: a census. The Semantic Substrate portfolio is, to our knowledge, the largest single classified corpus of agentic-web canonical strings — 4,017 domains, each hand-classified by architectural concept, layer position, moat structure, and industry applicability. Classifying it taught us how the agentic namespace is actually shaped. The numbers below are computed live from that classification, as of June 2026.

The namespace is concept-dense, not brand-dense

Of 4,017 classified names, only 181 are pure brandables. The rest are compound concept strings — two or more architectural primitives fused into a coordinate (attributionledger, orchestrationfabric, provenancegraph). The moat distribution makes the pattern explicit: 3,474 names (86%) carry exactly two concept moats, 358 carry one, and 185 carry three or more. The two-concept compound is the natural unit of the agentic namespace — one concept names the capability, the other names the surface it operates on.

Where the concepts concentrate

The ten densest concept clusters in the corpus:

• Attribution — 357 positions. The single largest cluster, and not by accident: attribution is the substrate primitive every higher-order trust concept (provenance, lineage, governance) grounds in.
• Orchestration — 317. The meta-category workhorse of multi-agent systems.
• Agent — 274. The era's namespace prefix of record.
• Intelligence — 215.
• Platform — 200 and Protocol — 190. The organizing-layer claim words.
• AGI — 171.
• Brandable — 181 (the non-compound exception).
• Substrate — 162 and Infrastructure — 158.

Concentration follows the trust stack: the deepest clusters are not product words but verification primitives — the vocabulary agentic systems will need to describe their own accountability.

The layer pyramid is inverted

Mapped to the four-layer architecture (L1 Substrate up to L4 Organizing, plus cross-cutting concepts that apply at every layer):

• Cross-cutting: 2,978 names (74%)
• Substrate (L1): 453
• Meta-category (L3): 316
• Organizing (L4): 141
• Meta-meta (frontier vocabulary — AGI, ASI, agentic web itself): 129

The organizing layer is the scarcest — there are only so many ways to say "the OS of X" — while the substrate layer (attribution, provenance, lineage, observability) is the deepest non-cross-cutting stratum. Scarcity at the top, density at the foundation: the namespace prices like real estate zoned by altitude.

TLD economics: .ai claims the category, .com anchors it

The corpus splits .ai 2,107 / .com 1,493 / .network 262, with long tails in .app, .cloud, .xyz, and .io. The more interesting number is the overlap: 523 matched root pairs exist where the same canonical string is held in both .ai and .com. A matched pair is a different asset than two singles — it forecloses the most common confusable, which is why the classification tracks pair membership as a first-class property.

Industry surface area

The corpus maps to 46 industries with explicit applicability tagging. The deepest vertical concentrations follow regulated and logistics-heavy domains — freight alone accounts for 150 concept positions — consistent with the thesis that attribution-class vocabulary matters most where accountability is already legally priced.

Method

Every name was classified on a master sheet with primary/secondary/tertiary category, an ordered moat list (compound surface), industry applicability, and derived layer position; counts here are computed from the live classification graph, not estimated. Layer definitions and the four-layer model are documented in the thesis; individual concept definitions live in the glossary. The portfolio itself is browsable at /browse.