When a beginner tells me they are confused by Salesforce’s AI names, I never correct them. They are right to be confused. Einstein, Copilot, Agentforce, prediction, generation, agents — the words pile up, they overlap, and some of them have quietly changed meaning over the past couple of years. The confusion is not a sign you are slow. It is a sign the branding genuinely moved faster than the explanations did.
So let me do what I always tried to do in a classroom: not impress you with the vocabulary, but give you one clear mental map you can actually keep.
Start with the three jobs AI does here
Forget the brand names for a moment. Underneath all of them, Salesforce AI has been doing three increasingly capable things over time.
First, it predicts. Given your data, it tells you what is likely — which lead will convert, which case might escalate. This is the oldest layer.
Second, it generates. It produces new content — a summary, an email draft, a reply suggestion — written in natural language.
Third, it acts. It does not just predict or write; it reasons about a goal and takes steps to accomplish it, using defined actions and real data.
Every name you are about to meet maps onto one or more of these three jobs. Hold that idea and the rest falls into place.
Einstein: the long-running brand
Einstein is the umbrella name Salesforce has used for its AI features for years. Originally it was mostly about prediction — Einstein scoring, Einstein forecasting, that kind of thing. As generative AI arrived, Salesforce extended the Einstein name to cover generative features too, like drafting emails or summarizing records.
So when you see “Einstein” attached to something, read it as: an AI capability built into a Salesforce feature. It is the predict-and-generate layer woven directly into the tools people already use. Einstein did not disappear when newer names arrived; much of it continues underneath them.
The “Copilot” idea: AI that assists you
For a while you also heard about a Salesforce “Copilot” — a conversational assistant sitting inside the application that you could ask for help. The useful concept here is assistance: AI that works alongside a person, answering questions and helping with tasks, while the human stays in the driver’s seat.
The name itself has been folded into the broader Agentforce direction, so do not get attached to the label. What matters is the pattern it represents: a helper you converse with. That conversational, in-app assistance lives on inside Agentforce today.
Do not memorize the names. Memorize the three jobs — predict, generate, act — and ask which one a given feature is really doing.
Agentforce: the shift to acting
Agentforce is the current center of gravity, and it represents the third job: acting. An Agentforce agent does not just suggest or draft. It reasons about what a user needs, chooses from a set of defined actions, and carries them out, grounded in your real data.
This is a genuine shift, not just a rename. Earlier AI mostly handed suggestions to a person who then did the work. An agent can do the work itself within the boundaries you set. That is why Agentforce comes with its own vocabulary — the agent, its actions, the planner that decides what to do, and grounding that ties answers to truth. If you have read my other posts, those four words are already familiar friends.
Agentforce does not throw Einstein away. It builds on it. The generative muscle that drafts an agent’s reply is Einstein’s generative layer at work; the agent simply adds reasoning and action on top.
A simple timeline to hold
Here is the map, in order of how the capabilities arrived:
Einstein began as prediction built into Salesforce features. As generative AI matured, Einstein grew to include generation — summaries, drafts, suggestions. The Copilot idea introduced conversational, in-app assistance. And Agentforce brought the leap to action — agents that reason and do, grounded in real data — absorbing the assistant concept along the way.
You do not need to track every product name Salesforce ships. You need to recognize which of the three jobs a feature is doing. When someone shows you a new “Einstein this” or “Agentforce that,” ask: is it predicting, generating, or acting? That single question will keep you oriented far longer than any list of names.
Where Data Cloud fits
One more piece worth knowing, because it sits underneath all of this. The quality of any AI — predictive, generative, or agentic — depends on the data it can reach. Data Cloud is what unifies customer data from across systems into one coherent view, and it is increasingly what gives Salesforce AI something rich and accurate to work from. The names on top may keep evolving. The need for good, unified data underneath them does not.
Your next step
To go deeper on the newest and most capable layer, read What Is Agentforce?. To understand the data foundation all of it rests on, read What Is Salesforce Data Cloud?. For more beginner-friendly guides, explore the Agentforce category.