Agent accuracy in production
Most agents have a memory now. Far fewer are accurate - able to answer correctly, stay right as things change, and show where each fact came from. Agent accuracy is the gap between a memory layer that simply exists and one you'd trust in production.
Most agents have a memory now. Far fewer are accurate - able to answer correctly, stay right as things change, and show where each fact came from. The whole field is circling this: Anthropic pulled vector search from Claude Code, "forgetting" became a design goal, and OWASP added memory poisoning to its 2026 agentic top ten. Agent accuracy is the gap between a memory layer that simply exists and one you'd trust in production.
In this webinar, Tobie Morgan Hitchcock (Co-Founder & CEO, SurrealDB) breaks agent accuracy into four properties you can actually build and measure - precise recall that fuses signals instead of betting on embeddings, correction that supersedes rather than overwrites, provenance on every fact, and the knowledge-graph structure that keeps retrieval exact. Anchored in one production scenario, end to end - with the benchmarks behind the claims.
In this session you'll learn
- What "agent accuracy" actually means - and why present and accurate are two different problems.
- The four properties that make memory accurate: recall, correction, provenance, and structure.
- How to measure whether your own memory layer is accurate, or just present.
