Salesforce Dreams Of The Agentic Enterprise
The management issue is not whether agents are possible; it is who owns them, how they are controlled, and how value is proved. For CIOs and transformation leaders, the immediate challenge is less model quality than the operating model: which business areas are allowed to deploy agents, who approves exceptions, and what guardrails stop duplicated automations from proliferating across functions and vendors.
That makes portfolio discipline essential. Agent initiatives should be treated as a constrained investment pool, not a broad capability bake-off. Leaders will need explicit criteria for prioritising use cases: process repeatability, data readiness, regulatory exposure, and measurable business pain. The trade-off is speed versus control; moving fast with one vendor can create later integration and lock-in costs, while a wider multi-vendor stance can slow delivery but reduce strategic dependency.
Governance also needs a new scorecard. Traditional AI enthusiasm is not enough if there is no way to track adoption, exception handling, human override rates, cycle-time reduction, or risk incidents. Ask whether the organisation has a clear owner for agent policy, a review cadence for retired or duplicated automations, and a benefits baseline before scale-up. Without that, โagenticโ becomes a label applied to experiments rather than a managed capability.
The practical next question is enterprise design, not product selection. Which capabilities must be standardized centrally, and which can remain embedded in business units? How will security, auditability, and data stewardship work when agents span CRM, ERP, and external platforms? Those decisions will determine whether agents become a durable operating advantage or another layer of unmanaged technical debt.
The dust has settled on Dreamforce, Salesforceโs well-attended San Francisco fete. Our takeaway? Salesforce is setting the pace with a platform for AI agents embedded in business applications, and itโs made huge progress on its products and partnerships in the last year. Salesforce claims that its (newly renamed) Agentforce360 product has Original Postress-releases/2025/09/03/fy26-q2-earnings/">6,000 paying customers. But in customer conversations and sessions, we saw little adoption or impact from AI agents. Lots of potential, but a long way to go for a meaningful RO.
Salesforce threw around the term Agentic Enterprise but failed to define it. We think they mean a Salesforce customer that ubiquitously uses AI agents defined by Forrester as AI applications tuned to act on behalf of an enterprise or individual, performing tasks, making decisions, and interacting with data or other systems autonomously.ย Agentic AI according to Forrester refers to systems that enable AI agents to operate, coordinate, and be governed within enterprise or digital environments.ย
Hereโs our take on Dreamforce 2025 for existing and potential enterprise customers. Salesforce has:
- Fixed its branding, thank goodness. Salesforce, with its โdigital laborโ drumbeat, was one of the original sinners in telling CEOs and boards of director that AI = People, so who needs people? Regardless of CEOs pounding their shoes on the table, managers and employees actively resisted using work-provided agents. Who would use a thing designed to take away their job? Itโs also balderdash โ AI agents are power drills, not people โ only people know where to drill and why. In an impressive feat of company transformation, Salesforce CEO Marc Benioff fixed the branding snafu and re-energized his company by walking away from the job-destroying sound of AI as โdigital laborโ to the more inspiring โAI elevates human potential.โ ย We think this positioning frames agents for what they are โ power tools that make your employees and customers more successful.
- Made a sales pitch for your enterprise โ not just your CRM โ agents. Salesforce has doubled down to build out the Agentforce product to be an enterprise-grade agentic platform. We like this for your CRM applications, but Salesforce is not stopping there and aspires to be your enterprise agent platform. They are not the only ones โ ย business software providers, hyperscalers, data platforms, iPaaS players, search, AI platforms, ITSM and ERP vendors โ all aspire for that land grab. ย We like Salesforceโs security model (which masks PII and other protected data when making a call to a language model), Amazon hosting for scale (and a deal with Google for hosting coming), Slack providing a coordination layer (to connect and activate โdata and dialogueโ), and a consumption-based pricing model option. We advise enterprises to start with targeted sales, marketing or customer service agents with Agentforce before attempting full-scale automation projects across functions such as supply chain. No surprises that the Dreamforce keynote cases studies focused on narrow use cases, lacked details on metrics, ROI, or measure reduced burden on HITL cognitive load.
- Showcased commitment to AI agents, but core product innovation was incremental. Thereโs no doubt where the โno softwareโ company sees its (and your) future: Enterprise AI, not just software. It renamed its entire portfolio from cloud to Agentforce (e.g. Marketing Cloud to Agentforce Marketing). It goes beyond marketing โ from what we saw and heard, Salesforce has been focused on expanding AI capabilities for each of their products and unifying products on the platform to make them AI Agent ready. Many other software companies are following a similar strategy but Salesforceโs advantage is the hundreds of packaged AI agents for their core and industry clouds. Enterprise users will need to choose between leveraging those OOTB agents but give up on the breadth beyond SF or go with a vendor that has limited prebuilt agents but spans into other apps to avoid lock-in.ย Innovation on core products was not center-stage.
- Doubled down on the data layer. AI agents without proprietary data and knowledge to ground them in your enterprise reality are just consumer-grade tools used at work. The key difference is grounding your AI agents in your data (which you know and already fund) and your knowledge assets (which lie scattered in file systems, repositories, and documents with ambiguous stewardship). Salesforce aims to solve that by asking you to move all your knowledge and data into its newly renamed Data 360 product. This strategy will likely bring much more accuracy and capability for the enterprises who do begin to shift their data gravity โ but in the agentic world, Agentforce must play well with others or be left out of the game. We donโt see Data 360 replacing Databricks or Snowflake vector databases. But we do like the informal positioning of Data 360 by a senior product strategist as the agent grounding card catalog rather than the knowledge repository.
- Announcements were mixed on how MuleSoft will enable the agentic enterprise. Agent Script is a good step towards bringing deterministic outcomes through rules and outcomes โ however it appears limited for complex tasks. We were disappoint that there was no similar announcement leveraging MuleSoft iPaaS to handle more complex scenarios. In contrast, MuleSoftโs Agent Fabric announcement was quite interesting. This leverages the API management side of MuleSoft to bring cataloging, governance, agent orchestration, and runtime visibility across all agents โ even those outside of Agentforce. A control tower to govern your portfolio of agents across vendors will be necessary.ย To be a truly agentic enterprise, Saleforce must recognize that Enterprise AI strategy will be multi-vendor and any orchestration platform must address multi-vendor agent sprawl.
Salesforceโs vision for an Agentic Enterprise is ambitious and timely. But ambition alone wonโt guarantee success. Customers need clarity on architecture, governance, and cost, not just pricing and a pragmatic roadmap to success. And most of all they need evidence that AI agents pay off in a proper return on investment.
Interested in connecting with one of us about Salesforceโs strategy? Clients can reach out to schedule a call with our analysts atย [email protected].
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.