Summary

Späti provides a primer on three layered concepts — semantic layers, context layers, and ontologies — arguing they are complementary rather than competing tools in the agentic era. He identifies a central paradox: agents both create the need for richer context and are the only thing that makes managing that context feasible at scale. The post distinguishes semantic layers (deterministic, SQL-compiled metric definitions) from context layers (unstructured business knowledge fed to reasoning agents) and ontologies (world models for cross-domain entity relationships). He cautions against defaulting to sophisticated retrieval (graphs, embeddings) when a flat index often performs better, framing sophistication as a cost that must earn its keep. The conclusion is that despite rapid agent capability growth, the hard problem remains governance — deciding what's true and keeping it maintained still demands human judgment.

Key Insight

Semantic layers, context layers, and ontologies are complementary tools with different cost-benefit profiles suited to different retrieval and consistency problems — but the persistent hard problem is governance, and deciding what's true still requires human judgment that agents cannot yet replace.

Spicy Quotes (click to share)

• 6

  It's a paradox: if we didn't have agents, we wouldn't need this much context. But we need more context because of agents, and we can only manage it because we have agents.

• 6

  Sophistication is a cost you should only pay if the outcome is worth much more than that cost.

• 7

  Someone still has to decide what's true.

• 5

  The hard part in data is never building semantics or the context once, it's how and who keeps it correct and maintained.

• 5

  An agent can read a table's schema, but the schema won't reveal the inside knowledge of how 'revenue' is defined, which tables join to which, or which columns are unreliable.

• 7

  So before reaching for a graph, an ontology, or a deterministic interface, ask what the retrieval or consistency problem actually is, who consumes the result, and what a wrong answer costs. Match the tool to that answer, not to what's trendy this quarter.

• 4

  Large language models are getting exceptionally good at writing the SQL with the right context, and at maintaining just that. But most often, deciding what 'revenue' means still requires deep company-specific knowledge (for now).

Tone

educational, analytical, opinionated, practical