The Biggest Agentforce Release Yet
Salesforce has begun rolling out Summer 26, described as its largest Agentforce update to date, in production waves starting June 13 with general availability on June 15. The headline capability is the graduation of multi-agent orchestration from beta to general availability, a step that lets specialized agents operate as a coordinated team across channels, sharing context to complete end-to-end workflows that span inquiry, quote, order, and after-sales service. It is the difference between a single helpful assistant and an orchestrated workforce of them.
Multi-agent orchestration matters because the real complexity of enterprise work rarely fits inside one agent's scope. A customer request often touches sales, fulfillment, and support systems, and a single monolithic agent struggles to handle that breadth well. By letting purpose-built agents hand work to one another while preserving shared context, Salesforce is aligning its product with how enterprise processes actually flow. The capability moving to general availability signals confidence that it is ready for production, not just demos.
Putting a Google Model Inside Its Own Platform
The most strategically revealing element of the release is the addition of native support for Google's Gemini 3.5 Flash, chosen for its speed and cost characteristics in the high-volume calling that agentic workloads demand. Salesforce embedding a competitor's model inside its flagship AI platform is a clear statement of where it believes its advantage lies. An enterprise agent platform, as Salesforce's own positioning makes plain, does not make one model call; it makes enormous volumes of them as agents reason, retrieve data, coordinate, and act.
That volume changes the economics. When a single business task triggers many model invocations, the cost and latency per call become decisive, and a fast, inexpensive model like Gemini 3.5 Flash becomes attractive precisely for the unglamorous, high-frequency work that agents perform constantly. By treating the model as a swappable component rather than the crown jewel, Salesforce is betting that its defensibility comes from owning the customer data, the workflow, and the trust layer, while the model underneath remains a contested, commoditizing input.
The Moat Is the Data and the Workflow
This is the through-line of Salesforce's entire AI strategy, and Summer 26 makes it explicit. The company is wagering that in a world where capable models are increasingly available from many providers, the durable advantage belongs to whoever sits closest to the enterprise's data and embeds most deeply in its daily workflows. Salesforce has both: decades of accumulated customer data and a platform that millions of users already work in every day. The model is rented; the position is owned.
We think this is a clear-eyed reading of where the industry is heading. Models are improving and converging, and exclusive access to any single one is an eroding advantage. What does not commoditize easily is the proprietary data, the established workflows, and the trust relationships that determine whether an enterprise will let an agent act on its behalf. Salesforce is organizing its product around the parts of the stack that resist commoditization, and embedding a rival's model is the most candid possible expression of that priority.
Analytics Agents Get a Direct Line
Summer 26 also includes Tableau MCP, a secure, open integration that lets agents query Tableau's analytics engine directly while operating under the Agentforce Trust Layer. This is a meaningful capability rather than a checkbox. Connecting agents to a mature analytics engine means they can ground their reasoning and actions in actual business data and visualizations rather than operating on whatever happens to be in their immediate context. Decisions made on real analytics are categorically better than decisions made on guesses.
Routing that access through the Trust Layer is the part enterprises should appreciate most. Letting an autonomous agent query analytics is powerful and potentially risky, and doing it under a governance layer that enforces permissions and oversight is what makes the capability deployable rather than merely impressive. The combination of analytical grounding and governed access reflects a maturing understanding of what agents need to be useful in production: not just intelligence, but trustworthy, controlled access to the systems where the truth lives.
What CIOs Should Take Away
For technology leaders, Summer 26 reinforces a planning principle worth adopting regardless of vendor: architect for model optionality. Salesforce's willingness to embed Gemini 3.5 Flash shows that even the largest platform providers now treat the model as a replaceable component, and enterprises should structure their own AI investments the same way, avoiding deep dependence on any single model when the landscape is shifting this quickly. Optionality at the model layer is cheap insurance against a fast-moving market.
The release also clarifies what to scrutinize when evaluating agent platforms. The differentiators are no longer raw model benchmarks but the quality of orchestration, the depth of data integration, and the rigor of the governance layer. Those are the capabilities that determine whether agents can be trusted to act in production, and they are where the real engineering difficulty lies. Salesforce is competing on exactly that ground, and its customers should evaluate it, and its rivals, on those terms rather than on whose model tops a leaderboard this month.
