Salesforce Ships Agentforce Memory Update Aimed at Doubling Agent Recall
The Development
Salesforce published an update to Agentforce Memory on November 4, 2026, under the title "Double recall: Inside the Agentforce Memory update." The headline claim is a doubling of recall, which points to agents retaining and surfacing substantially more prior context during customer interactions. The Salesforce Newsroom post was the only source available, and its body was not accessible at the time of writing, so benchmarks, methodology, availability tiers and pricing remain unverified. What is confirmed is the release itself, the product (Agentforce Memory, part of the Agentforce platform) and the positioning: recall is now a headline performance metric for Salesforce's agent stack, not a background feature.
Our Take
The "double recall" framing matters more than the number, which we cannot yet verify. Salesforce is turning memory into a measurable product claim, and that shifts the competitive axis for agents from model quality to context quality. Models are converging; the proprietary customer record is not. An agent that remembers a customer's last three service cases, purchase history and stated preferences without a retrieval workaround is only possible where the agent sits on the system of record. That is Salesforce's structural advantage, and recall is how it monetises it. Treat the claim as directional until the methodology is published, but plan on the assumption that memory depth will be a procurement criterion by mid-2027.
What Changed
Agents can now carry materially more customer context between sessions and across tasks. Persistent, expanded recall turns Agentforce from a stateless responder that re-asks questions into a continuity layer that remembers preferences, prior issues and commitments, which is the precondition for agents owning multi-step customer journeys.
Marketing Impact
CRM, lifecycle and service-adjacent marketing teams feel this first. Agents with deeper recall can sustain personalised follow-up without rebuilding context per touchpoint, cutting the manual briefing and prompt-stuffing that currently limits agent autonomy. Marketing operations inherits a new governance task: deciding what agents are allowed to remember and for how long.
Competitive Implication
Platforms that own the customer record gain, because memory depth compounds with proprietary first-party data that standalone agent vendors must rent through integrations. Point-solution agent startups and horizontal copilots without native CRM access are vulnerable: they can match model quality but not persistent, permissioned context at the source.
Strategic Outlook
Expect HubSpot, Adobe and Microsoft to answer with their own memory claims within two quarters, and recall benchmarks to become a standard line in agent RFPs by mid-2027. Privacy and retention controls will follow as the second battleground, since longer memory expands consent and data-governance exposure.
The Exploit
Action Item
CMOs and heads of marketing operations running Agentforce should have their Salesforce account team deliver the recall methodology and retention controls before Q1 2027 planning closes, then pilot one memory-dependent journey, such as renewal or win-back, against a stateless control.