Agentforce is hitting the ecosystem like a storm, and so far, it’s looking to be one for the books. Having worked in the Salesforce sphere for the better part of a decade, I’ve seen my share of launches and re-namings, but Agentforce has a different feel to it. For once, it seems that Agentforce is worth its hype – it can fundamentally alter how Salesforce is used and for what. It’s no wonder then that Salesforce themselves, along with ISVs and partners, are going full throttle to bring Agentforce to the market.
Agentforce represents Salesforce’s new product strategy, built on dynamic usage-based pricing, over the tried-and-true licensing model. Gone are the days of indirect, often obscure business benefit calculations. Agentforce taps into core business processes like sales development and shopper concierge services, providing direct value and bottom-line impact. While I see Agentforce as a positive development, there is also a lot of misunderstanding around Salesforce’s new approach. In this article, we cover five of the most common misconceptions surrounding Agentforce.
1. Agentforce is Just a Re-Packaged Einstein Copilot
A typical misconception I’ve heard is that ”Agentforce is just a renaming of Einstein Copilot” or ”just another AI chatbot”. This is simply not true. Agentforce is part of a wider paradigm shift called ”agentification”. In this new paradigm, AI is given agency to perform tasks on behalf of humans. This means that AI models can not only interpret natural language and respond in kind but also act upon it.
The interesting part is that Agentforce is able to trigger actions and identify relevant field values from user prompts. How does Salesforce make this happen? Agentforce uses a technology called Atlas Reasoning Engine to identify topics and parse natural language prompts into a workable JSON format that Salesforce can use to launch Flows, Apex, and prompt templates.
Another question is what is driving the agent’s actions. The (not-so) secret sauce behind Agentforce is Flow. It’s already the one-stop shop for low-code automation (excluding Omnistudio), so it makes sense that it’s the primus motor of Agentforce. Essentially, Agentforce uses natural language instead of record updates or external API calls to trigger premade actions.
With Agentforce, Flow becomes an even more versatile and powerful automation tool. Flows will take additional roles that have typically been left to human reasoning, like placing the correct order or activating the right case resolution.
As a result, the number of Flows will likely increase as more of the previous human activity gets run by AI. The flip side of this is that the large number of these ”mini-flows” used by Agentforce places increasing challenges for admins and architects to orchestrate the processes at a coherent level.
2. Agentforce is Super Expensive to Use
Would you pay $2 for a chatbot discussion? Probably not. But this is a terrible comparison in reality. An autonomous agent should always be benchmarked against a human doing the same work. How many conversations can a sales rep handle in an hour? What about their hourly wage and indirect costs, like training and software licenses? Account for these and the $2 per conversation is exceeded sevenfold.
Agentforce is only expensive if you apply it to the wrong business use case. Since the pricing model is usage-based, what you should look at is unit margins. You want to ensure that each conversation (on average) leads to direct cost savings or revenue generation exceeding $2.
This means that as volume scales upwards, you’re making more and more money with Agentforce. However, a word of warning: the mechanism works the other way around too. If you place an agent in the wrong one, you’re bleeding money. Agentforce will only be expensive if the use case doesn’t net more revenue than the conversation cost. That’s why use case planning is so vital for Agentforce.
The bottom line is that you shouldn’t view Agentforce as an investment; think of it more as a virtual employee. Let’s say you have a process that needs staffing. You can look at several alternatives: insourcing, outsourcing, process automation — and autonomous/semi-autonomous agents. An investment would be a fixed cost that you pay to get dividends over time.
3. Agentforce Won’t Work Without Implementing Data Cloud
First, a disclaimer: Data Cloud is, technically, a prerequisite for Agentforce, since it is used behind the scenes for the Einstein Trust Layer, conversation audit logging, and data processing. That said, a full implementation and paid Data Cloud credits are not required for Agentforce. You can run Agentforce by simply “flipping the light switch” on for Data Cloud and leveraging the zero-dollar SKU that Salesforce Enterprise Edition customers get for free.
What does “full implementation” then mean? Data Cloud is a versatile powerhouse of a data platform. It can power customer data consolidation, real-time identity resolution, data enrichment, and segment activation within Salesforce and external systems.
To enable all this, you need to integrate data sources, map them to data model objects (DMOs), build identity resolution reconciliation rules, etc. This all takes time, effort, and skill to achieve, leading to a hefty implementation price tag. To top it off, data ingestion, sharing, and activation all consume Data Cloud usage credits.
“Provided you have the necessary data and assets in Salesforce, nothing is preventing you from firing up an agent — with or without Data Cloud.” Salesforce
With the minimum implementation of Data Cloud in use, you can still build and run AI agents, but they lose the ability to leverage external data consolidated by Data Cloud. This means no access to unified profiles, calculated insights, or attributes included in related DMOs. In addition, unstructured data from e.g. pdf files or emails is not available for agents, since Agentforce uses Data Cloud for retrieval augmented generation (RAG).
Make no mistake – Data Cloud is a great asset for grounding LLMs in Salesforce. However, full implementation of Data Cloud is not a prerequisite for Agentforce. You can still provide context for an agent with the conversation itself, Salesforce records and metadata, CMS files and documents, and knowledge articles.
Provided you have the necessary data and assets in Salesforce, nothing is preventing you from firing up an agent – with or without Data Cloud.
4. Agentforce Only Works if Your Data Is in Salesforce
It’s no secret that Salesforce wants to be involved with as much of your data as possible. In the past, this meant that Salesforce required you to replicate your ERP or OMS data as Salesforce objects and set up bi-directional sync to ensure proper data mastering. An alternative has been the use of external objects to surface data from an external system without having to replicate it. Data Cloud has added a third data integration option with its zero-copy approach to data sharing.
Regardless of the option you go for, the goal is to make data available in Salesforce. But what if you can’t or don’t want to bring data into Salesforce? There are valid business reasons for this, such as data residency, cost of ingestion (as is the case with Data Cloud and external objects), or vendor lock-in avoidance. Even if this is the case, Agentforce ways of working around these limitations.
The good news is that with the help of Flow, you can set up an API call without coding. For the AI agent to access external data, it needs to activate a Flow to call the external system. For this, you need an autolaunched Flow that has an action step to make an HTTP callout to the external system. An important caveat to note: you don’t necessarily need to know Apex or other coding languages for this, but experience with APIs is required.
5. Agentforce Is Worthless if Your Data Is Bad
This one is a half-truth. Yes, your data sets the foundation for Salesforce AI usage, but keep in mind that Agentforce has means for filling in the blanks. For one, there’s a whole foundational model behind the agent that has years of training and refinement behind it. Even without data to provide context, you have the full capability of LLMs like GPT-4 accessible via the Salesforce UI and protected by the Einstein Trust Layer to avoid data leakage.
Secondly, Salesforce fields are not the only source of truth for Agentforce. Combined with Data Cloud, it uses RAG to retrieve knowledge in files and documents. Data Cloud is capable of tapping into unstructured data by applying natural language processing to information in PDF files or emails.
This is made possible by Data Cloud’s Vector Database which makes unstructured data easily retrievable and analyzable by LLMs. Essentially, Data Cloud is able to skim through countless documents to search for relevant information to provide answers and drive actions.
Furthermore, companies who say that their data is no good still use rule-based automations that rely as much on data as AI solutions. We sometimes forget that Flows and Apex triggers have been around for a much longer time than generative AI, and data quality issues aren’t anything new either.
It’s better to assume that data quality won’t be perfect, and build that into your agent workflow. For instance, you can include disclaimers in agent instructions to highlight uncertainty in data reliability.
Agentforce can also be a part of the solution. A universal truth about CRM systems is that no one enjoys manually creating records or typing field values. This is where Agentforce comes in with its ability to summarize and populate Salesforce fields on behalf of users.
Agents can be a valuable asset when it comes to ensuring both data consistency and completeness. As a rule of thumb, in terms of data quality, the further you can minimize manual data entry, the better.
Summary
Agentforce platform is easy and quick to configure but setting up customised agents takes more time and effort. Think of agent building as an opportunity cost to employee or partner onboarding. In that sense, you should weigh the costs and benefits of agents vs. other means of tackling a business process. Agents have their limitations, and they need to be positioned correctly to yield the most business value.
The best way to find out Agentforce’s value for your business is to start exploring. After all, as a new platform, Agentforce’s documentation is still under development. The best way (for now) is to spin up an agent and test it out.
To get a first glimpse into the possibilities of Agentforce, the best first step is to complete the quick start Trailhead module. If you want to take your exploration a step further, I suggest joining or organizing an Agentforce hackathon.