Source: STLabs — Graph Over Tables (Mar 2026)

STLabs’ March 12, 2026 engineering post, by Sam Makarovskiy, is the clearest public explanation of Axiom as a context graph rather than a conventional CMDB.

Key Claims

• Traditional CMDBs fail because they are systems of record that humans must maintain; if updates are not a side effect of actual work, data decays ^[extracted].
• STLabs designed Axiom around three requirements: zero manual data entry, relationships as first-class objects, and source merging without losing provenance ^[extracted].
• A graph database matches service-management questions because blast radius, dependencies, ownership, access, and impact are naturally relationship-shaped ^[extracted].
• STLabs argues vector search is useful for semantic lookup, but insufficient for structural questions like “what is affected by this network change?” ^[extracted].
• Axiom combines vectors for semantic understanding with graph traversal for structural reasoning ^[extracted].
• Before an agent runs a workflow, it can walk the graph to understand downstream risk; without the graph, the agent is guessing ^[extracted].
• Every graph fact is a timestamped assertion with source and confidence. Example shape: (Laptop-X) --[assigned_to]--> (Alice) with source, observed_at, and confidence metadata ^[extracted].
• Identity resolution keeps native IDs queryable even after merging records across systems like Jamf, CrowdStrike, Intune, Okta, and Workday ^[extracted].
• The post explicitly acknowledges graph trade-offs: harder to operate than Postgres, less mature tooling, smaller talent pool, and risk of bad graph modeling ^[extracted].

Why It Matters

This post is stronger than generic “AI ITSM” marketing because it names the data-model bet. For initlabs, the implication is that context-graph credibility needs more than a diagram: buyers may expect provenance, identity resolution, structural traversal, and a clear explanation of when vectors are insufficient.

Limitations

• The post explains the architecture, not production evidence.
• It does not disclose the graph database, traversal language, exportability, scale benchmarks, or customer deployment data.
• It does not prove that the current product exposes the described provenance and graph traversal UX to buyers.

Promotes Into

• context-graph
• stlabs
• research-stlabs-competitor