Agentforce reached general availability as agents that take action across sales workflows, and Einstein Copilot was folded into it months later.
The distinction Salesforce is drawing — agents that act versus assistants that suggest — decides who is accountable when the output is wrong. That is a live question for any team letting software touch an opportunity record.
A copilot suggests; a person accepts or rejects. An agent decides and acts within limits you set. Salesforce's position is that the second is where the value is, and it renamed its assistant product accordingly.
In practice an agent here is a configured workflow with a model in the loop — qualifying an inbound lead, updating a record, drafting and sending a follow-up — running without a human clicking each step.
The platform underneath is low-code, which means the people configuring these agents will often be admins rather than engineers — a governance fact as much as a technical one.
The short version: software that does the step rather than proposing it, and Salesforce has bet its AI branding on that difference.
Salesforce unveiled Agentforce on 12 September 2024 and made it generally available on 29 October 2024, describing agents that take autonomous action across sales, service, marketing and commerce.
General availability: 29 October 2024.
Named use cases: Resolving customer cases, qualifying sales leads, optimising marketing campaigns.
Naming: Einstein Copilot, introduced April 2024, was renamed Agentforce in January 2025 with no functional change beyond permissions, interface and documentation updates.
Price: Not stated in the announcement.
Guardrails: Not detailed — the announcement does not specify the controls limiting agent action.
What it doesn't do: the announcement does not explain what constrains an agent, which is the only question that matters before granting one write access to your pipeline. The rename also means anyone searching internal documentation for Einstein Copilot is reading about a product that no longer exists under that name.
SDR — The most exposed role, and it is worth being direct about that. Autonomous lead qualification is explicitly one of the named use cases, and it targets work that currently justifies SDR headcount.
AE — Mostly upside near-term. Agents handling record updates and routine follow-up removes admin without touching the judgement work that closes deals.
Manager — Your accountability model needs revisiting before rollout. When an agent qualifies a lead out and the deal was real, the post-mortem needs an owner, and today most teams have not named one.
RevOps — This is a configuration and guardrail project, not a switch. Agent permissions, escalation paths and audit trails are the work — and the Einstein Copilot rename means your existing documentation is now wrong.
Before delegating qualification to software, write down the standard you would hold a person to. If you cannot state it, an agent cannot apply it.
Two minutes, once a week. What changed in AI, and what to run because of it.