AI Specialist: What I learned taking the certificate?
Salesforce AI
This post is not about materials you can study to pass the certificate. There are plenty of information out there by Salesforce Ben, Focus on Force, Apex Hours to prep for that.
It is more about what I understood about AI, how Salesforce is leveraging AI with its product and what they are trying to sell. So, let’s dive in:
What did I learn?
I have been writing Prompts all wrong
I had an eye opening revelation about prompt writing. So far I've been giving LLMs simple instructions to generate content. Those would either hallucinate or give too vague answer.
Once I did few trailheads, I realized there’s a method to writing good prompt. A good prompt needs to include: a good instruction + contextual information + Constraints + specific direction.
Here’s an example:
Look at the level of detail above! If it was for me, I would have just said:
‘Write me a blog post about cake decorating techniques’
I highly encourage my readers to go through this trailhead module: link.
Salesforce is selling trust with AI
You may have heard of Einstein Trust Layer. If not, it is the whole concept of how Salesforce protects your company’s data when it is interacting with external LLMs like ChatGPT, Claude 3, Gemini Pro.
So, what Salesforce is selling is this trust layer which will effectively protect your company’s confidential data (for example, customer’s phone number, social security, company’s revenue).
It is doing that by 3 main tools:
Dynamic Grounding: Anchors responses in your org's contextual data in a secured way
Data Masking: Masks sensitive information before external LLM processes
Zero Data Retention: Imagine a Mission Impossible-style message that self-destructs after being read by LLMs
This picture will help:
Salesforce is selling a virtual assistant
If you are in Salesforce ecosystem, you must be tired by now of hearing Agentforce. What is it really?
Think of it like a chatbot but on massive AI-powered steroids.
Chatbots is a dumb tool that provides limited response to the questions you ask.
While Agentforce is an assistant who can perform certain actions based on questions/requests you pass.
Marc Benioff recently talked about how in the movie Minority Report, Tom Cruise is assisted by a virtual agent to select clothes for him. Link here.
This agent is the concept behind agentforce, whose role is to assist humans in completing certain tasks.
Let’s take 2 Salesforce examples:
As a Sales rep, your primary task is to manage a customer’s Account. But, instead of doing that manually you want to hire a skilled intern who can:
Generate customized client emails
Log calls in your account record
Create and manage a to-do list
Set up automated reminders
Now that skilled intern is Agentforce.
As a Salesforce admin, you want to find duplicate Contact records in your org. You should be able to use Agentforce to identify a list of such records based on Name, email, create a report based off of that and maybe someday just merge the records once you review it.
Currently, you may do that using some app exchange dedupe tool. In future you can just write prompt for Agentforce.
Functionally, you achieve that in Salesforce by:
First define the Topics or roles the Agentforce will do, scheduling an appointment, managing an order, escalating a case.
Next step, define an Action: list of tasks to perform, using apex class, flows or prompts, (like identifying available time for appointment and booking an appointment).
Watch this video too: link
So essentially what Agentforce is doing is removing bunch of manual steps that you would otherwise do or hire someone else to do.
Salesforce is selling an assistant who can hopefully save you time and increase your productivity.
Great article!