Salesforce has shipped a lot of AI brands in eighteen months. Einstein, Einstein Copilot, Agentforce, Agentforce Vibes, and the original predictive features still underneath all of it. Here is what each one actually is, what it is for, and how to pick between them without getting lost in the naming.
Salesforce has renamed its AI portfolio often enough that even people who work on the platform full time sometimes lose track. In the last eighteen months we have seen Einstein GPT, Einstein Copilot, Agentforce, Agentforce 2.0, Agentforce Vibes, Agentforce Vibes 2.0, and assorted sub-features underneath each. The product has moved faster than the vocabulary.
This is a quick guide to what each brand actually is today, what it is for, and how to pick between them when you are scoping work. No hype, no keynote slides.
A short history, because it explains the naming
Einstein is the original Salesforce AI brand, going back to 2016. For most of its life, Einstein meant predictive features: lead scoring, opportunity scoring, case classification, forecasting. Narrow models doing narrow things. Some of this still exists and still works well.
In 2023, Salesforce started bolting generative AI onto the platform and called it Einstein GPT. That became Einstein Copilot. At Dreamforce 2024, Salesforce repositioned the whole thing and called it Agentforce. In 2025, Agentforce got a developer-focused sibling called Vibes. In 2026, Vibes 2.0 arrived alongside Headless 360.
So when someone says "Einstein" today, they could mean the legacy predictive features, the generative features that were briefly called Copilot, or the overall AI brand. The context usually makes it clear. Usually.
What each thing actually is in 2026
Einstein (the legacy brand)
Still there. Still useful. Predictive lead and opportunity scoring, Einstein Activity Capture, Einstein Case Classification, Einstein Forecasting, Prediction Builder. These are narrow ML models that sit inside the platform, trained on your data, producing scores and classifications on records.
If you want to enrich records with predictive signals (who is likely to convert, which cases are urgent, what an opportunity is likely to close at), Einstein in the legacy sense still does the job well. It is not generative. It is not an agent. It is a prediction service.
Agentforce (the current umbrella for agents)
Agentforce is the platform for building and running AI agents. Customer-facing or internal. It includes the agent builder, the large language model connectivity, the skills library, the Atlas reasoning engine, and the governance tools.
When we say "agent" here, we mean: a bounded AI actor that can take actions inside Salesforce (create records, run flows, call Apex, invoke integrations) in response to natural language input, with guardrails on what it is allowed to do. Think of it as a layer above Flow and Apex that understands intent and routes it to the right capability.
Agentforce is where the rubber meets the road. If you are building a support agent that resolves tier-one cases, a sales agent that drafts outreach, or an internal operations agent that answers questions across multiple systems, Agentforce is the product.
Agentforce Experience Layer
Newer, and worth knowing. This is the component of Agentforce that renders rich interactive UI (cards, workflows, forms) across surfaces like Slack, Teams, ChatGPT, Claude, and Gemini. It separates what an agent does from where it is rendered, which makes multi-surface delivery practical.
Agentforce Vibes (and Vibes 2.0)
Vibes is the Salesforce-aware AI for developers. It is not an agent you run for customers. It is a development partner that lives inside the IDE (or in Agentforce Studio) and helps engineers build on the platform.
Vibes understands your org. It can generate flows, scaffold custom tabs, write Apex, review LWC, and deploy changes. The 2.0 release added multi-model support (Claude Sonnet 4.5 as the default, GPT-5 available), plan and act working modes, and deeper org awareness.
If someone on the engineering team is asking "how do I build this Salesforce feature faster," the answer is often Vibes. If someone on the business side is asking "how do I automate this customer interaction," the answer is Agentforce.
The decision tree
A simple set of questions that resolves most scoping conversations.
Are you building for customers, partners, or employees who are not developers?
This is Agentforce. Pick a use case, scope the agent, configure the skills and guardrails, deploy. You are likely paying per-conversation or per-user, depending on the edition.
Are you accelerating your Salesforce engineering team?
This is Vibes. Install it, connect it to the org, and the development team uses it for daily work. Pricing is per-seat for developers.
Do you want to enrich records with predictive scores?
This is legacy Einstein. Predictive scoring, classification, forecasting. No conversation, no agent. Just a model producing numbers on records.
Do you want an agent embedded in Slack, Teams, ChatGPT, or another external surface?
This is Agentforce plus the Experience Layer. The agent lives in Agentforce. The Experience Layer is how you ship it into the client.
Do you want your agents calling custom business logic you already have?
Agentforce can call Apex methods, invoke flows, or hit external APIs as "skills." The work is mostly in defining those skills clearly and deciding what the agent is and is not allowed to do.
Common confusions
"We want to use Einstein for our chatbot." Usually, you want Agentforce. "Einstein" in the legacy sense was not a chatbot platform. If someone sold you an Einstein chatbot engagement, check what product they actually mean. It might be Einstein Bots (the older bot platform), which has been largely superseded by Agentforce.
"We want Copilot."Copilot was renamed to Agentforce. You want Agentforce. If someone is still selling you "Einstein Copilot" in 2026, they are a year behind.
"We want Vibes for our users." Vibes is for developers, not end users. You probably want Agentforce.
"We want the agent to just work." All agent projects need scoping, skill definition, guardrails, and testing. The product lets you move fast, but an agent with no boundaries is the fastest path to a bad incident. Budget for the design work.
Pricing notes
Pricing on Salesforce AI features has moved around and we are not going to quote current numbers here, because they will be wrong by the time you read it. A few principles that tend to hold.
Agentforce is usually billed on a consumption basis (per conversation or per action) in addition to platform licenses. Budget for the variable cost based on expected volume.
Vibes is typically per-developer seat, and the math usually works out if developers are using it daily. The hourly productivity uplift has to cover the seat cost.
Legacy Einstein features are bundled into various edition tiers. Check which ones are included before paying extra for Einstein add-ons.
What is actually worth doing first
If you are new to Salesforce AI and trying to figure out where to start, here is what we would suggest.
- Turn on legacy Einstein scoring features if your edition includes them. Free lift for most sales and service orgs.
- Pick one Agentforce use case with a narrow scope, clear guardrails, and measurable success criteria. Ship it, measure it, iterate.
- Give Vibes to two or three engineers who are receptive to working with AI tooling. Let them report back in a month.
Three small bets, three different kinds of value, no platform-wide transformation program. You will learn more than from a twelve-month strategy exercise.
The brand will change again
It probably will. Salesforce is still figuring out what to call all this, and the naming will keep evolving. The underlying capabilities are real and the direction is consistent. Focus on what the thing does, not what it is called this quarter.
If you need help picking between them for a specific project, or you want to talk through whether an agent use case is worth the scoping effort, that is a conversation we have about twice a week. Happy to add yours to the list.