Agentforce Blueprint: Salesforce’s 5-Step Guide to Building Digital Employees

What exactly is a digital employee, and how do you build one? Digital employees are the next step in the evolution of AI assistants, from multi-purpose chatbots to identities that can perform some tasks in roles traditionally held by humans.

Salesforce is all in on this concept. At the Salesforce Financial Services Summit in Sydney this week, the software giant spoke more about Agentforce, the company’s new framework for creating digital employees that can plan, reason and act autonomously within business systems. (Salesforce launched a version with agents customised for financial services organisations.)

Salesforce has a structured five-step process for building a digital employee. Here's how they’re rethinking productivity in financial services businesses with AI agents.

‍

What Is a Digital Employee?

A digital employee, according to Salesforce, is an AI agent capable of reasoning, planning and executing tasks without human intervention.

This goes beyond the typical AI features seen in apps over the past few years – such as flagging unusual transactions or transcribing calls – which detect patterns and carry out basic actions. What sets agents apart is their ability to plan. That means processing multiple layers of context: industry knowledge, company-specific processes, and the specifics of the task itself.

Planning with context is a key improvement. Increasingly, these agents will also be able to execute multiple tasks in sequence.

The goal is not just smarter tools, but something more akin to a digital teammate. An autonomous system that understands its role, has access to the necessary data, and can take meaningful action across systems, with or without human involvement.

‍

The Five Assets of a Digital Employee

Salesforce outlines five assets as guiding principles for building and managing a digital employee. These principles help organisations design these AI agents with clear responsibilities, appropriate access and the ability to take action without making obvious mistakes. 

At the launch event, Keran Wijetunga, director, AI and data solution engineering at Salesforce, gave five guiding principles for building and managing a digital employee:

1. Role
   What is the job the agent needs to perform? Whether it's assisting customers directly or supporting internal teams, a well-defined role is the foundation for functionality.
   “We start with the job and the responsibilities of that job,” said Wijetunga. “Is this going to be a customer-facing agent helping resolve transactions, or maybe an internal agent supporting employees?”
2. Data Access
   What data is needed to do the job, and where is it located? Agents must connect to both structured (e.g. CRM systems) and unstructured (e.g. email) data sources.
   “You need to connect those agents through to your structured and unstructured data, but also to your metadata and your semantics – the data that describes your data,” Wijetunga said.
   
   
3. Action Capabilities
   What can the agent do? Organisations can expose pre-built workflows, APIs and automation tools to agents so they can take action—moving from reactive to proactive.
   Organisations don’t need to rebuild the automations they have already made, Wijetunga said. They just have to give the agent access to those automations so it can do its job.
4. Guardrails and Governance
   What should the agent not be able to do? Defining limits is just as important as enabling capabilities, ensuring the agent acts within policy and compliance boundaries.
   “It’s probably just as important to define what an agent can’t do than what it can,” Wijetunga said. “We need to respect the principle of least privilege.”
5. Deployment Channels
   Where will the agent operate? It could be through email, Slack, mobile apps, Salesforce, or all of the above. The deployment context needs to reflect where work actually happens.
   
   

What is Agentforce? Salesforce’s Bet on Agentic AI

To turn these principles into working systems, Salesforce has built Agentforce, a platform to build and manage digital employees. Rather than piecing together fragmented tools, Agentforce is a single, integrated architecture that contains the essential components for intelligent, autonomous agents.

“An LLM by itself doesn’t drive business value,” Wijetunga said. “It’s like having an engine without the roads and tunnels and bridges. You need the rest of the system.”

The system combines:

• LLMs (Large Language Models) for understanding and generating language.
• Data Cloud for real-time access to structured and unstructured data. This uses “zero-copy integration” to access information in its existing location rather than copying it to another storage pool.
• Financial Services Cloud as a collection of use cases and industry-focused templates, with predefined roles, APIs and use cases.
• A Reasoning Engine that mimics how a human plans, evaluates and refines actions – asking itself “do I have enough information?” before executing a task.
  
  

These components are not standalone. They sit within a broader architectural model that Salesforce says is essential for agentic AI to work in real business environments. Underpinning Agentforce is a four-layer structure that matches each capability to a specific operational foundation:

• Data – Context-specific, business-grade information
• Apps – Operational systems where tasks are performed
• Agents – The digital employees themselves
• Governance – Controls for privacy, permissions and auditability
  
  

This layered approach addresses three of the most common failure points in enterprise AI deployments. 

First, by grounding agents in reliable, context-rich business data, it reduces the risk of hallucinations – plausible but incorrect responses that can undermine trust and utility. 

Second, integrated governance controls enforce data access policies and compliance obligations. This avoids the security and privacy gaps that can emerge when AI systems operate without guardrails. 

And third, by embedding agents within the application and automation landscape, the architecture ensures scalability. It enables agents to perform consistently across different processes, user roles and interaction channels as demand grows. 

“Data, apps, agents, and governance – these are the four key ingredients for an agentic AI system to ensure that it’s effective, that it’s scalable, and that it’s trusted,” said Wijetunga.

‍