Dreamforce has consistently showcased Salesforce’s thinking on the future of enterprise technology. This year, the focus was on extending its platform to support AI agents across applications and interfaces, with data, context, and governance at the centre.
On the ground at Dreamforce, Ecosystm Founder and CEO Amit Gupta looked beyond the product announcements to the thinking behind them: Salesforce’s leadership, culture, and long-term direction. Ecosystm Analysts Sash Mukherjee and Darian Bird examine the announcements, demos, and messaging, highlighting what stood out and what deserves closer attention.
The Organisation Behind The Technology
Dreamforce offered a perspective on Salesforce that goes beyond its products and technology: the organisation Marc Benioff is building around them. From discussions with the broader team, one thing became clear: great founders are unconditionally committed to what they build, and that conviction is evident in how he leads Salesforce. His ability to build relationships and alliances with the most relevant and influential leaders in the sector, combined with what his leadership team describes as his “beginner’s mind”, helps create a culture that continues to question what works rather than rely on past success.
As AI reshapes the technology landscape, this founder-led approach helps a large global player such as Salesforce stay anchored in relentless innovation, supported by strong partnerships and a venture arm, Salesforce Ventures, that gives Salesforce and its customers early exposure to emerging innovations.
Benioff’s belief in doing good while doing well is embedded in Salesforce’s culture through its 1-1-1 model, which dedicates 1% of equity, product, and employee time to community causes. Salesforce employees have collectively contributed 10 million volunteer hours, supporting causes including education, youth development, and environmental conservation. The model has also inspired the Pledge 1% movement, which now includes more than 19,000 companies committing resources to social impact.
The same principle extends to responsible technology. In 2018, Salesforce appointed a Chief Ethical and Humane Use Officer and established a dedicated office to embed ethical considerations in product development and AI use. This is a function that should play a significant role in helping boards at Salesforce’s largest customers assess the ethical, organisational, and business implications of their AI roadmaps.
Salesforce Ventures reflects another dimension of this forward-looking approach. With investments in more than 700 companies, including Anthropic, it gives Salesforce early exposure to technologies that could complement its offerings and create new opportunities for customers. Investing from seed through growth stage also gives the company a view of emerging innovation well before it reaches the mainstream. Importantly, it is not primarily a pipeline for acquisitions: by its own account, Salesforce has acquired fewer than 5% of the companies it has backed. The emphasis is on long-term, patient capital rather than on feeding Salesforce’s M&A agenda.
This broader perspective is important as AI agents begin to orchestrate work across applications, including through interfaces such as Claude. Salesforce is anticipating a future in which it may not own the customer interface, but instead provide the data, context, and execution capabilities that underpin experiences delivered through other platforms. This is a forward-looking view of its role in AI-driven work, but it also changes the basis of customer loyalty.
As the interface becomes less central to the relationship, Salesforce’s stickiness will depend more on the trust it has built with customers, its data platform, its industry relationships, its approach to responsible technology, and its ability to anticipate where innovation is heading. These strengths could help Salesforce remain integral to how organisations use AI, even when customers engage with its capabilities through other platforms.
Salesforce’s approach of extending beyond its own capabilities and drawing on the wider innovation ecosystem, while maintaining a focus on responsible and ethical AI deployment, could prove a powerful combination for a company with its breadth of enterprise-grade capabilities.
Salesforce’s Enterprise AI Strategy
1. Salesforce extends its platform beyond its own interface
Salesforce’s central architectural message at Dreamforce was the Enterprise AI Harness, a layer connecting models to enterprise data, context, workflows, permissions, actions, and controls. Its applications provide the customer records, business rules, permissions, and processes that agents need to operate within an organisation.
According to Salesforce, these capabilities should remain accessible even when the user interface belongs to another provider. A sales manager working in Claude, for example, could retrieve account history or initiate an authorised Salesforce workflow, with Salesforce providing the underlying information and executing the action.
- AIforce. An interface layer bringing Salesforce data, business logic, permissions, and actions into environments including Claude, Slack, and Salesforce itself.
- Headless Toolkit. A set of development tools exposing Salesforce capabilities through APIs, MCP servers, skills, and plug-ins, allowing organisations to build tailored interfaces without rebuilding the underlying logic.
- Claudeforce. An extension of the Anthropic partnership bringing Salesforce into Claude through a plug-in with prebuilt sales skills and Salesforce MCP connectivity. Authentication and permissions remain managed by Salesforce, which also executes the actions.
Ecosystm Opinion
Opening Salesforce to external AI interfaces could extend its reach beyond users who work directly in its applications. The opportunity is to retain a central role in enterprise workflows as other providers take on more of the user interaction. Its established data, business processes, and integrations give it a strong position, but that position will depend on the value customers continue to derive from these capabilities, regardless of where they access them.
The commercial model will be an important factor. External agents are expected to be metered through Flex Credits, although pricing multipliers have not been announced. As AI-mediated interactions grow, customers will need to assess these costs against the benefits of making Salesforce capabilities available across more interfaces. Salesforce has an opportunity to expand its role in enterprise AI, but the economics of external access will help determine how readily customers embrace it.
2. Agentforce moves from assistants to digital workers
Salesforce is positioning Agentforce to execute defined business processes, extending beyond answering questions or generating content. Agents retain context, monitor progress, perform authorised tasks, and escalate issues requiring human judgement. Salesforce outlines three stages for realising ROI: deploy quickly, customise and extend, then observe and optimise against business outcomes.
- Long-horizon agents. These agents pursue goals over time, retaining memory across sessions, preserving plans between executions, adapting as they learn, and responding to user feedback. Hunter, Salesforce’s AI outbound sales agent, illustrates this approach by monitoring at-risk deals over days or weeks and adapting follow-up plans to re-engage prospects.
- AgentScript & specialised agents. AgentScript lets organisations define agent behaviour, including deterministic routing and delegation to specialised agents. Salesforce’s Multi-Agent Orchestration capability allows a higher-level agent to coordinate specialised agents across a broader process.
Salesforce reported that Fulton Bank deployed Agentforce Coworker in six weeks to support 3,000 employees, saving 80,000 hours and generating USD 389 million in new loans and deposits.
Ecosystm Opinion
Salesforce has a strong foundation for applying agents to business processes with measurable outcomes. Its established presence in sales and customer service gives a natural role in supporting operational tasks. Stable, high-volume processes with clear completion criteria offer a practical starting point, and the Fulton Bank deployment illustrates how quickly these capabilities can be introduced across an organisation.
Results will depend on the work being automated, its integration with external systems, and how exceptions are handled. Hours saved may reduce expenditure or free employees for higher-value work, depending on how organisations use that capacity. Service quality, intervention rates, and the total cost of implementation, consumption, and ongoing oversight provide a fuller picture of the impact.
Longer-running agents could extend automation to processes that unfold over days or weeks. Their performance will depend on how well they respond to changing circumstances, with checkpoints and defined human review points helping manage exceptions. Evidence from sustained deployments will show where agents can deliver reliable outcomes with an acceptable level of oversight and cost.
3. Trusted data & context become foundations for agentic work
Salesforce’s position is that reliable agentic work depends on data that is accessible, authoritative, consistently defined, and supplied with relevant business context. A customer record alone may not tell an agent what a metric means, which source to trust, or how current the information is. Metadata, business definitions, lineage, customer history, and real-time signals help provide that context.
- Data 360 & the Agent Context Engine. Designed to provide agents with customer history, business definitions, permissions, activity, and real-time signals as needed. Salesforce cited potential token-cost reductions of up to 30 times.
- Informatica’s AI-ready data approach. Informatica’s framework assesses readiness across seven factors and 21 measures, with the lowest-scoring factor determining the overall result. This highlights specific weaknesses rather than masking them in an average score.Announcements also included master data management integration with Data 360 and catalogue integration with MuleSoft Agent Fabric.
- Zero-copy access. Data 360 is designed to make governed data available across systems without routinely duplicating it into another repository, potentially reducing data-pipeline requirements. Integration, latency, and cost will still depend on the source systems.
Ecosystm Opinion
Combining Salesforce’s customer and process context with Informatica’s capabilities could extend its agentic proposition beyond CRM. Decisions in finance, supply, and operations often rely on data held across multiple systems, making consistent definitions and reliable access important. Informatica could help connect Salesforce’s capabilities to these data environments, particularly for customers using multiple clouds and applications.
An agent recommending payment terms, for example, needs to know which source to use when billing and CRM records disagree. Organisations must establish authoritative sources, reconcile inconsistencies, and keep information current as business conditions change.
Token savings are only one part of the economics. Sending less information to a model can reduce consumption, but preparing, retrieving, governing, and maintaining context also requires resources. Organisations will need to weigh these costs against the reliability of agent outputs and the effort required to connect existing data sources. The aim should be to make existing data usable for agentic applications without requiring a wholesale rebuild of the data estate.
4. Slack & Tableau bring collaboration, insight & action together
Salesforce is bringing agents and analytics into the same workflows where employees discuss issues, make decisions, and take action. Slack is becoming the collaborative interface for interacting with agents, data, and applications, while Tableau is extending analytics from reporting towards proactive detection and action. The objective is to reduce the distance between an insight, the discussion around it, and the business process that follows.
- Slack as an agent and application interface. Slackbot and agents operating in shared channels allow employees to interact with AI alongside colleagues, using conversation as part of the working context. Slack Code and other agent capabilities extend this model beyond CRM, allowing agents to work within collaborative channels rather than separate applications. Review, testing, and security controls become important as agents move from providing information to taking action.
- Tableau as operational analytics. Tableau is adding natural-language analysis, proactive insights, and governed business definitions through Tableau Agent and Tableau Semantics, with Agent Actions able to trigger automated workflows. Salesforce cited Mastercard growing from 100 to 1,000 Tableau users in months while embedding agentic analytics.
Ecosystm Opinion
Bringing agents and analytics into Slack could connect discussions, analysis, and action within the same work environment. For Salesforce, this extends access to its data and AI capabilities beyond employees who work directly in its applications. Adoption will depend on whether teams find it more useful than their existing collaboration tools and AI interfaces. Slack’s success will depend on earning a central role in how employees work, rather than relying solely on Salesforce’s installed base.
As agents contribute to conversations and trigger workflows based on analytical findings, organisations need ways to review outputs and intervene where necessary, particularly where customer commitments or financial outcomes are involved.
5. Governing agents becomes an enterprise management requirement
While inaccurate information retrieval can often be addressed through stronger checks, agents that incorrectly update customer records, issue credits, or initiate processes can have a far-reaching impact. Salesforce’s position is that this makes agents a new class of actor, requiring security and governance beyond conventional application controls. Security must work in both directions: protecting the AI systems an organisation deploys and protecting the organisation from what those systems do.
- Identity & permissions. Agents require their own identities and appropriately scoped permissions rather than broad human credentials, covering access to data, systems, and actions.
- Agent Fabric & observability. MuleSoft Agent Fabric provides capabilities for registering and discovering agents, enforcing policies, managing identity, and monitoring activity across agent ecosystems. Salesforce’s Observability Center provides broader monitoring of agent health and activity, while Agentforce has separate session-tracing and analytics capabilities. An Agent Kill Switch provides a mechanism for containing rogue agents.
- Broader security architecture. Salesforce Guardian brings together data security, privacy, and resilience capabilities, while new agent-focused capabilities address identity, observability, and security controls.
Ecosystm Opinion
Salesforce is extending its security capabilities to address the risks associated with agents that can act across business systems. Its opportunity is to make governance part of the infrastructure customers use to deploy and manage agents, with clear accountability for consequential actions.
Enterprises may use agents from several vendors alongside internally developed systems, so Salesforce’s controls will need to work with external identity, security, and monitoring tools. Customers will need clarity on which agents and actions are covered and where additional mechanisms are required. Salesforce could strengthen its position in enterprise AI governance by extending its controls beyond its own products to provide consistent oversight across this mixed agent environment.
Organisations will also need a basis for deciding when agents can be given greater autonomy. Evidence of their performance and the consequences of their actions should inform those decisions.
