A subtle UX failure in search systems: semantic retrieval finds a document about “A15 Bionic chip benchmarks” when you search “iPhone performance,” but keyword highlighting finds nothing to highlight because the exact words “iPhone” and “performance” don’t appear. The user sees a long document with no highlights and asks “Why is this here?” The retrieval was intelligent; the display was not.
The fix — semantic highlighting using small specialized models (0.6B parameters, millisecond inference) — points to a broader principle: every layer of the user experience should reflect the system’s actual intelligence level. This connects to The UI moat collapses — API quality becomes the purchasing criterion — if the UI can’t communicate why a result is relevant, the intelligence behind the retrieval is invisible to users. It also matters for The three-layer AI stack: Memory, Search, Reasoning: the Reasoning layer should surface its reasoning, not just its conclusions. For agent-native products described in In agent-native architecture, features are prompts — not code, the agent should explain its retrieval path, not just return results.