Blog 2
Getting started: how to build an AI agent in Salesforce?
In our previous blog, we gave an overview of what Agentforce is and what you can use it for within your company. But how do you specifically get started? In this post, we explain step by step how to set up your first AI agent within Salesforce with Agentforce.
You don’t need to be an AI specialist to do this. However, it is important to think about your agent’s purpose, what data it needs and what tasks it needs to perform.
Step 1: Define your agent’s purpose
Start with the question: what do I want to automate or support with AI? Preferably start with a limited and clearly defined process. Some examples:
- Automatically summarising customer conversations
- Giving suggestions on incoming customer queries
- Answering internal questions about procedures or products
- Look up data in your CRM and display it in understandable language
The more concrete the goal, the better the agent can function.
Step 2: Set your agent’s properties
Within Agentforce, each agent is built according to five characteristics:
- Role: what is the agent’s job?
- Knowledge: what may the agent draw from? (e.g. customer data, product information, internal guidelines)
- Actions: what is the agent allowed to do? (give answers, create tasks, make summaries…)
- Channels: through which channel does the agent work (e.g. chat window on your website, Slack, email, Teams…)?
- Boundaries: what restrictions are there? (e.g. no access to confidential files)
This structure ensures that your agent does not “just do anything”, but works within clear parameters.
Step 3: Use the Agent Builder in Salesforce
Salesforce provides a user-friendly Agent Builder that lets you build an agent step by step. This is a low-code environment, which means you don’t need any programming knowledge, but you do need to think logically about flows and responses.
Through that builder, you can include:
- set up prompts
- selecting data sources
- linking actions to buttons or triggers
- testing how your agent responds to questions
A well-built agent is not only smart, but also predictable and safe to use.
Step 4: Test and improve
Before you put an agent “live”, test it in a secure environment. Let several users work with it and collect feedback. Adjust where necessary: maybe it doesn’t understand certain questions, or you need to provide an extra data link.
AI is not a “set and forget”. Most organisations are constantly refining their agents based on usage.
Step 5: Integrate into your operation
An agent is only useful if it is actually used. Make sure your team knows where the agent sits, how it works and what it is for. Give short training sessions, create an internal manual if necessary and link its use to your existing processes.
Example: if an agent automatically creates call reports, you should also effectively teach employees where to find those reports and how to reuse them.
In conclusion: start small, think big
Agentforce’s strength lies in its flexibility. You can start with one simple agent, and then expand to other departments or applications. By working step by step, you keep control and build knowledge.
In the following blog, we show some practical examples of agents at Flemish SMEs: what they do, what benefits they bring and how employees interact with them.
