[object Object]

What It Does

Agentic Enterprise Search (Spring 2026) unifies search, collaboration, and action in one interface. Powered by Data 360. Context-aware across 200+ external sources — SharePoint, Google Drive, Notion, Confluence, Box, Jira, ServiceNow Knowledge, GitHub, Zendesk, and more. Coordinates actions across multiple AI agents through Agent Fabric, so a query that surfaces a draft contract can also kick off a Legal-team review action without leaving the search result. Output format is conversational, with cited source chunks and one-click pivot to the full document.

The Problem It Solves

Knowledge lives everywhere. Reps and agents copy-paste between tools, lose context, and rely on the muscle memory of “I think I saw it in Notion.” Agentic search provides one query interface — get the answer, launch an action, collaborate on the outcome, all without leaving Salesforce. The retrieval architecture combines BM25 lexical search, dense vector search via Data 360 embeddings, and a re-ranker that boosts results by recency, source authority, and the asker’s role-based access.

Query: "What's our policy on customer-supplied AMIs?"
  Sources hit:  Confluence (Security wiki), SharePoint (Legal contracts),
                Salesforce Knowledge (Service runbooks)
  Top result:   Confluence page (last updated 12 days ago, 4 citations)
  Suggested action: "File a Security Review request" (via ServiceNow Action)

Setup Considerations

Connector coverage varies in depth. SharePoint and Google Drive are first-class and indexed in near-real-time; smaller-tail sources sync hourly or on demand. Index freshness matters — stale knowledge returns stale answers, and the freshness lag is configurable per connector. Permissions must pass through correctly via SCIM, OIDC, or per-source service principals, or users see content they shouldn’t. Audit the connector’s security model before enabling in production, and run a permission-leak test using a low-privilege test user before broad rollout.

Connector setup checklist:
  [ ] Authentication mode (OAuth, service principal, SCIM)
  [ ] Permission propagation tested with low-priv user
  [ ] Sync cadence matches content velocity
  [ ] PII redaction policy applied
  [ ] Source attribution formatting matches brand voice
  [ ] Connector cost reviewed (per-document, per-call, or flat)

Limits

Semantic search quality depends on embedding model and content quality. Garbage in, garbage out — a dump of 10-year-old PDFs degrades the index more than it helps. The “200+ sources” line looks impressive on a slide; most orgs connect 5–10 real sources that practitioners actually use daily. Focus on those. Latency is real: the 95th-percentile end-to-end query lands at 2.5–4 seconds with 8 sources active, which is fine for a thoughtful question and too slow for a tab-completion use case.

Use Agentic Enterprise Search for cross-source questions, for agent-driven workflows that need grounded context, and for any user who already lives in Salesforce. Stick with native source search for single-source deep dives (a Confluence-only knowledge audit, for example) where you want the source’s full faceted UI.

Common Failure Modes

  • Indexing every source on day one. Start with 3 high-value sources, validate, then add.
  • Skipping permission audits. The first leak destroys trust for a year.
  • Treating cost as zero. Embedding storage, query compute, and connector API quotas all add up.

What to Do This Week

Pick the three knowledge sources your reps and agents query most often, configure their connectors, and run 20 representative queries to validate answer quality and permission scope before broadening.

[object Object]
Share