Orchestration lessons from Salesforce, Zendesk, Hilton
Recent writeups show Salesforce reframing agent reliability as an orchestration issue, Zendesk evolving with layered observability and audit trails, and Hilton launching an AI Planner that acts as a cross‑brand orchestration layer — all stressing orchestration, incremental rollout, and traceability. ( )
Salesforce formalized a "six levels of determinism" for Agentforce — Level 1 through Level 6 — naming "Agent Script" as the top level for enforcing deterministic, script-driven actions inside agents (Salesforce authors: Olfa Kharrat, Reinier van Leuken). (salesforce.com) Bertrand Duperrin’s March 23 analysis notes Salesforce is explicitly shifting execution responsibility toward scripted orchestration layers, and warns that adding deterministic scripting increases implementation cost and complexity for customers. (duperrin.com) Zendesk’s March updates detail concrete agent capabilities that enable traceability: generative replies that draw from the knowledge base, multi-intent detection that can combine articles and execute multiple workflows in one email, and a new email integration option to send AI replies as a system user (freeing an agent license). (premiumplus.io) Zendesk operational controls include an account-level audit log that retains a full change history for Enterprise/Enterprise Plus customers and developer APIs for actions, events, and escalation flows that let teams capture context and programmatic traces for handoffs. (support.zendesk.com) Hilton launched the Hilton AI Planner into beta on March 10, 2026 at hilton.com, exposing the feature to a small pool of site traffic and positioning it as a conversational, cross‑portfolio planner with real‑time curated recommendations, per CIO Michael Leidinger’s announcement. (stories.hilton.com) Across the three vendors the operational pattern is concrete: Salesforce prescribes layered determinism and an "Agent Script" top layer, Zendesk exposes audit logs + developer APIs to capture decision lineage and escalations, and Hilton is using a small‑traffic beta to validate conversational planning across brands before broader rollout. (salesforce.com)
