Salesforce renamed the Einstein Copilot for Salesforce agent type to Agentforce in the Spring ’25 release. That explains the older assistant name. The wider Agentforce family now covers more capabilities, so a buying decision still needs the specific use case, feature and entitlement.
The short version: Einstein Copilot for Salesforce became Agentforce in Spring ’25 — that agent-type rename did not change its functionality. This does not make every Agentforce product equivalent to the old assistant. Permissions, UI labels, and the help docs were all updated to match. The distinction people are usually reaching for when they ask this question is real, but it sits inside Agentforce: assisted work, where a person asks and the AI answers, versus autonomous work, where an agent acts on its own.
We have left this post at its original URL because the question still gets asked constantly, and a straight answer is more useful than a dead link. Below: what actually changed, what the surviving distinction is, where the Einstein name still appears, and how to sequence a rollout.
What actually happened to Einstein Copilot
Einstein Copilot launched as the conversational assistant embedded in the Salesforce UI — ask a question in plain English, get a record summary, a drafted email, or an answer without hunting through tabs. In the Spring ’25 release Salesforce folded it into the Agentforce brand. Nothing was switched off and nothing needed migrating; the product you were using kept working under a new label. Salesforce's own release notes are blunt about it: “Einstein Copilot for Salesforce is Now Agentforce.”
Then at Dreamforce in October 2025 the umbrella widened again into Agentforce 360, which introduced broader Agentforce branding. Current product pages again use Sales Cloud and Service Cloud, while release documentation can retain Agentforce labels. Our product-name guide distinguishes those labels from partnerships and interface layers. If you are reading vendor material that still compares “Copilot vs Agentforce” as a purchasing decision, it was written before 2025 or by someone who has not checked since.
The distinction that does still matter: assisted vs autonomous
Here is the useful part. People asking “Copilot or Agentforce?” are almost always asking a good question in outdated vocabulary. What they want to know is whether the AI should help their team work faster, or do the work without their team. Both live under Agentforce now, they cost different amounts, they carry very different risk, and they should be sequenced rather than chosen between.
| Person-initiated assistance | Autonomous agents | |
|---|---|---|
| Who starts it | A person asks | A customer interaction, event or schedule starts work |
| What it produces | Answers, drafts or authorised actions; configure review points | A completed action — record updated, case resolved, reply sent |
| Human in the loop | At the review or confirmation points you configure | Only at escalation points you define |
| Main risk | Incorrect answers or actions a user may trust | Incorrect actions repeated before monitoring catches them |
| Setup effort | Scope access, configure actions and train users | High — scope topics, wire actions, define guardrails, shadow-test |
| Time to value | Depends on scope, access and acceptance checks | Depends on actions, integrations and acceptance checks |
| Cost driver | Per-user licensing | Consumption — roughly $2 per conversation |
The row that catches people out is “main risk”. Assisted AI fails visibly — someone reads a bad summary and rewrites it. Autonomous agents fail quietly, and you find out from a customer. That difference, not the licensing, is why we push hard on shadow-mode testing against real historical cases before an agent is allowed to act.
So is the Einstein name gone?
Not entirely, which is the reason the confusion persists. The conversational assistant became Agentforce, but Einstein branding survives on the older predictive and scoring capabilities — lead and opportunity scoring, prediction building, and the analytics features that pre-date the generative era. If a Salesforce rep says “Einstein” to you today, they most likely mean predictive scoring rather than the assistant. Worth asking rather than assuming.
How to sequence it
Our recommendation has not changed with the branding: start assisted, then automate. Turn on the assisted capabilities for a team that lives in Salesforce daily, let them get used to checking AI output, and watch which requests they make over and over. Those repeated requests are your agent backlog, discovered from real usage instead of guessed at in a workshop.
Then pick one narrow, high-volume, low-blast-radius workflow for your first autonomous agent. Tier-1 case deflection and lead follow-up are the usual candidates because they are repetitive, measurable, and recoverable when the agent gets one wrong. Prove the pattern there before pointing an agent at anything that touches revenue or compliance.
The biggest mistake we see is treating this as a technology project rather than an operating change. The agent is the easy part. Deciding what it is allowed to do, who owns it when it misbehaves, and what “good” looks like in a dashboard is the work.
The bottom line
There is no Copilot-versus-Agentforce decision to make, because there is no Copilot to choose. There is a real decision about how much autonomy to hand over and how fast, and that one deserves genuine thought. Start assisted, earn the right to go autonomous, and be suspicious of anyone selling you agents before they have asked what your data looks like.
Working out where your org sits? Book a free Salesforce AI assessment with TechParrot and we'll map out the right approach for your org.
Working out where Agentforce actually fits?
Assisted and autonomous are different projects with different timelines. Tell us the use case and we will say which one you are really asking for.
Four fields, no call required. We reply within one business day.
Frequently asked questions
Is Einstein Copilot the same as Agentforce?
Salesforce renamed the Einstein Copilot for Salesforce agent type to Agentforce in the Spring '25 release without changing its functionality in that rename. The wider Agentforce family includes more capabilities, so confirm the specific feature and entitlement rather than treating every Agentforce product as the old assistant.
Does Einstein Copilot still exist in 2026?
Not under that name. The conversational assistant is now part of Agentforce, which itself sits under the Agentforce 360 umbrella introduced at Dreamforce in October 2025. Any material presenting Einstein Copilot and Agentforce as competing products predates the change.
Is the Einstein brand gone from Salesforce entirely?
No. Einstein branding survives on the older predictive and scoring capabilities, such as lead and opportunity scoring and prediction building. What moved to the Agentforce name was the generative conversational assistant. If someone at Salesforce refers to Einstein today, they usually mean predictive scoring rather than the assistant.
What is the difference between assisted and autonomous Agentforce?
Person-initiated assistance starts with a user request and can include answers, drafts or authorised actions. More autonomous workflows can start from customer interactions, events or schedules. Human review, permissions and escalation must be designed for the actual actions. Neither the label nor the trigger alone establishes the licence model, safety or rollout time.