Diagramming Agentforce Solutions
Agentforce changes the way we think about architecture.
Where traditional systems are designed, agents are trained, guided, and governed. This shift demands new ways of explaining how intelligence operates within our solutions.
The Diagramming Agentforce Solutions framework builds on the same principles as Diagramming Salesforce Solutions but extends them into the AI era.
Its goal is to help architects, designers, and leaders communicate how autonomous components create value, how they are controlled, and how they remain trustworthy.
At this early stage, the framework focuses on three essential diagram types introduced at Dreamforce 2025. These provide a starting vocabulary for visualising Agentforce systems, while hinting at a wider visual grammar that will continue to evolve.
Value Stream Mapping for Agentforce
Question: Where does AI create measurable value?
Purpose: To show how agent interventions improve process flow and customer outcomes.
This diagram uses a value stream perspective to compare the current journey with the agent-enhanced journey.
It measures improvements in lead time, process time, and handoffs, showing where automation and reasoning reduce friction.
Value stream mapping can be conducted for current and future states of the process. In the current state a swim lanes will most likely represent human actors, in the future state an additional lane is added for the AI agent, or agent. The supplied example is the future interaction flow variant where the AI agent is generically named "Agentforce", but this could be adapted to use the label applied to the agent, e.g., "Knowledge Agent".
By introducing an Agentforce swim lane, the diagram highlights the points where agents step in, enrich data, or take decisions, giving stakeholders a clear view of how value is created and verified.
Context & Retrieval Architecture
Question: What data feeds the agents?
Purpose: To describe how knowledge is retrieved, structured, and governed for reasoning.
The Context & Retrieval Architecture maps the information ecosystem that supports an agent’s decision-making.
It shows how Salesforce data, Data Cloud, and external sources combine through Zero-Copy Federation, retrieval pipelines, and embedding preparation.
It also captures the controls that protect data integrity, such as cleansing, de-duplication, and PII masking, and how decision traces and evidence are recorded for governance and compliance.
This diagram helps technical and non-technical audiences alike understand how the agent’s “worldview” is built and managed.
Trust & Governance Framework
Question:
What data is used, where it is governed, and how trust is assured?Purpose:
To show how every data source, process, and interaction in the Agentforce ecosystem is observed, governed, and auditable.
The Trust & Governance Framework illustrates the flow of data across three domains: Enrich, Interact, and Observe. It shows where structured, semi-structured, and unstructured data enter Data Cloud, how unstructured sources are processed through the Vector Embedding Process, and how Agentforce retrieves information securely using both Zero Copy federation and vector search.
The framework also highlights the observability layer, the logs, audit trails, and monitoring systems that provide continuous visibility into ingestion, retrieval, and AI reasoning. Together these controls ensure that every agent interaction with data is traceable, explainable, and compliant.
This diagram enables data, architecture, and compliance teams to share a unified view of how trust is enforced at every stage of the AI lifecycle, from data source to decision.
What Comes Next
The current release of the framework is only the beginning.
Future versions will expand to include Agent Interaction Flows, Training Data Architectures, Event-State Models, and Trust Layer Overlays, adding further precision to how we describe learning, reasoning, and collaboration among agents. Support for new MCP and A2A protocols will require diagrams to support additional details.
As Agentforce matures, so too will the diagrams that explain it.
For now, these first three serve as the foundation for understanding where AI acts, what it depends on, and how we can trust it to work on our behalf.