John Turner

<< Previous | Next >>

Daily Learnings: Tue, Jan 21, 2025

This world, after all our science and sciences, is still a miracle; wonderful, inscrutable, magical and more, to whosoever will think of it. — Thomas Carlyle

Answers to Agentforce Questions

The following are some technical questions and answers for the Salesforce Agentforce product that aren’t as easily found in documentation or other areas.

Does Agentforce “Learn”?

E.g., If a Salesforce AE says, “Agentforce learns continually to get better and better!”

Basically, at this time, the answer is “no”. Agentforce isn’t using machine learning, and it would potentially go against the 0 data retention policy. That said, it could learn and get better over time through continuous improvement and iteration on the topics, instructions, and actions that you configure. However, this isn’t an automated process (at this time).

Can Agentforce search / use unified knowledge articles coming from outside of Salesforce?

As of right now, no, but upgrades to the product coming potentially in Q2 (safe harbor) might be assisting with this. You could get around this through the use of Data Cloud and regular importing of the data.

I’m getting permissions errors in the Agent Builder for the SDR Agent, but I think that I’ve assigned all permissions needed. What’s happening?

The following profiles / permission sets are needed for the SDR agent:

Can I have an Agent start a conversation via Slack with someone based on the status of a record, and then update the record based on the response in Slack?

As of now, no. However (safe harbor) this might be coming with the forthcoming Agent API.

When uploading a FAQ PDF to an Einstein Data Library in my sandbox environment, I’m consistently getting an error that says “We couldn’t upload your file. Please try again.” But, this file uploads fine in a SDO. What is causing this?

This error isn’t something that Salesforce has seen before, and if you see this regularly, you should log a case with them. Though there isn’t a helpful error message, and since this works in the SDO, likely this is a permissions issue, rather than a file type or size issue.

Are there tools to help me better iterate on topics, instructions, actions, or the performance of my Agent?

At this time, no, but they’re working on a tool called Agentforce Conversation Explorer. This tool will group conversations into clusters, grade those clusters, and then provide insights into iterations on topics and instructions to assist with bettering agent configuration.

Can I provide a file to a conversation with an Agent for it to review and take action on?

Currently, no. But (safe harbor) this is coming. File support will enable Service Agents and Employee Assist with image and PDF processing capabilities. This will utilize LLMs and Data Cloud to extract the relevant information from the files.

When can we “build an agent from scratch” instead of using the pre-built templates?

Safe harbor - the target is for the March 2025 release.

When will the Sales Coach Agent be able to do the demoed “real time insights” feature, as it’s not currently possible yet?

Safe harbor - Real Time Conversation Insights will support 3rd party call providers starting with Zoom, followed closely by MS Teams and Google Meet. Date at the moment is TBD.

Agentforce pricing seems a little steep, can we use an in-house model instead to reduce costs?

At this time, no. You cannot currently adjust the model that the reasoning engine uses.

Is there a standard action for the Service Agent to allow them access to Salesforce Scheduler to suggest available time slots?

Not yet available.

Further question, when adding a file to the Data Library, how do I get the agent to find the data within the file? I heard that you cannot ground the Sales Coach Agent, is that possible?

Consider using CRM data via Flows, as well as sales methodology documents uploaded to Einstein Data Libraries.

There is a Salesforce help documentation page that says that you need to specifically use Prompt Builder, Einstein Data Libraries + Retrievers, and then wire up a custom action for each file that you upload.

I’m trying to use a Scheduled Flow to trigger field generation based on a prompt template built in Prompt Builder. I’m running into a limit error (see below). Are there different limits in a sandbox vs. in a production org?

Error: LLMGatewaySandboxAPICallsPerHour Limit Exceeded

Yes, there are limits depending on Partial, Full, and production orgs. You can reference the following article to review those limits.

References