Data 360 does not fix RAG with one switch—the benchmark must become a diagnosis
Salesforce reports a jump from 62.5% to 92.5% on knowledge-base questions after changing parsing, indexing, chunks, retrieval depth and the generation prompt. The result matters, but it becomes a production decision only when a team identifies where its own documents fail.
Read article
One agent or many? A new map for multi-agent architectures
SOMA separates complex domains across specialist agents while preserving one conversational front door.
Flow testing evolves to validate outcomes, not just paths
Assertions and test suites help admins prove that automation produced the expected state before production.
OrbTrail
Latest
31 stories found
Data CloudNewData 360 does not fix RAG with one switch—the benchmark must become a diagnosis
Salesforce reports a jump from 62.5% to 92.5% on knowledge-base questions after changing parsing, indexing, chunks, retrieval depth and the generation prompt. The result matters, but it becomes a production decision only when a team identifies where its own documents fail.
AgentforceNewClaudeforce moves CRM beyond the screen—but architecture still decides what can happen
Salesforce in Claude promises 37 sales skills, live data and governed actions inside Claude. The experience moves to a conversational interface, but identity, authorization, retention and operational evidence remain distributed across two platforms and an MCP path.
PlatformFlow’s new user context closes a trust gap—but identity still defines the boundary
Winter ’27 can pin screen and autolaunched flows to user context regardless of their caller. The option makes CRUD, field access and sharing predictable, but it does not replace permission sets, customer verification or persona-based tests.
PlatformConsent now crosses two Marketing Clouds. The source of truth still has to be designed
Email, SMS and WhatsApp synchronization reduces fragmentation between Marketing Cloud Engagement and Marketing Cloud Next, but identity, explicit opt-in and universal unsubscribe make deployment a governance project—not a simple list mapping.
PlatformSalesforce kept permissions in profiles. That does not make the profile an access architecture
Cancelling the retirement removes a migration deadline, not the scaling problem. A sustainable model leaves only profile-dependent settings in the profile, composes capabilities in groups, and automates grants, revocation and expiration.
AgentforceAmbiguous metadata makes Agentforce guess: remediation must end with a test, not a filled description
Field, object, Flow, topic and action descriptions provide operating context for an agent. The real challenge is repairing only what it can reach and measuring whether selection and outcomes improve.