Salesforce used its Dreamforce keynote on Tuesday to launch AIforce, a layer designed to let people and AI agents reach data, workflows and business logic held inside Salesforce through Anthropic's Claude, Slack and other AI interfaces.

The announcement, delivered by Salesforce CEO Marc Benioff, carried an unusual degree of supplier concentration for a single keynote. Anthropic's chief executive joined Benioff on stage, as did the CEOs of Siemens and NVIDIA, for a discussion of how AI agents could reshape corporate work.

The technical substance of AIforce lies in what it exposes. Salesforce systems hold customer data, process logic and operational workflows that, historically, required trained staff working through dedicated interfaces. AIforce opens that surface to external AI systems — Claude first among them, through an integration branded Claudeforce, with Slack and other AI interfaces also supported. For R&D organizations that run project pipelines, partner management or grants administration on Salesforce, the implication is direct: agents could query and act on those records without a human navigating the application layer.

The presence of three external CEOs alongside Benioff signals how the vendor landscape is consolidating around agentic AI as the next procurement cycle. Siemens brings industrial engineering and manufacturing workflows into the conversation; NVIDIA supplies the compute substrate on which frontier models and agent inference run; Anthropic supplies the models themselves. Salesforce positions itself as the system of record through which these capabilities reach enterprise data.

For R&D managers, the rollout raises questions worth tracking before any budget commitment. First, access control: an agent layer that can reach business logic inside Salesforce expands the attack and error surface, and Salesforce has not yet detailed in public materials how permissions map to agent-initiated actions. Second, governance: organizations in regulated research environments will need to know whether agent queries and outputs are logged to the same standard as human actions. Third, cost: inference-heavy agent workloads carry compute costs that scale with usage, a line item most R&D software budgets do not yet model.

The timing of the announcement coincides with what the host publication describes as a surge in digital R&D spending, suggesting Salesforce is positioning AIforce to capture a share of that growth. The keynote format itself — vendor CEOs presenting jointly rather than through standalone product launches — marks a shift toward bundled, multi-vendor AI offerings aimed at corporate buyers rather than individual tool purchases.

What remains unmeasured, at this stage, is deployment reality. Tuesday's keynote described capability; it did not report customer trial data, adoption figures or measured productivity outcomes from AIforce deployments. Claims about AI reshaping corporate work remain, for now, projections from the companies selling the stack. Research organizations evaluating the layer will want reference deployments, permission architecture documentation and cost-per-query benchmarks before treating AIforce as production infrastructure rather than a demo.

Watch for early adopter case studies and pricing disclosure in the weeks following Dreamforce, which will determine whether AIforce moves from keynote stage to procurement shortlist.