If your service team runs an Einstein Bot, you are closer to a working Agentforce service agent than you might think. Salesforce’s Upgrade Agent tool, now generally available and listed among the 21 announcements from Dreamforce 2026 (15 to 17 September), converts an existing Einstein Bot into an Agentforce Service Agent or Help Agent. It builds the new agent in minutes and, importantly, builds it alongside the old one, so the bot keeps answering customers while the team reviews and refines its successor.
That design choice matters more than the speed. Most chatbot replacements begin with a cutover date and a nervous weekend. This one begins with two systems running side by side and a team free to compare them on real conversations before anyone switches anything off.
The teams getting the most from the upgrade treat the tool as a head start rather than a finish line. Years of bot operation have left them with transcripts, flows, integrations and escalation rules that describe exactly what their customers ask for. This article covers what the tool does, what carries across, how long a considered upgrade takes and how to show that the new agent is earning its place.
What does the Einstein Bot upgrade tool actually do?
The Upgrade Agent tool, found in Agentforce Labs, reads an existing Einstein Bot and scaffolds a new Agentforce agent from it. It takes the bot’s intents and dialogs as its reference and proposes the topics, instructions and actions an agent needs to cover the same ground. The new agent is created as a separate record, so the original bot stays live and untouched. Teams can test the upgraded agent, adjust it and decide when to move traffic across, rather than flipping a switch and hoping. Salesforce moved the tool from beta to general availability in the run-up to Dreamforce 2026, which means it is supported for production use today.
It helps to read the word upgrade generously but accurately. Einstein Bots and Agentforce are built differently, so the tool produces a strong first draft of the new agent rather than a line-by-line copy of the old bot. That is good news in practice. The parts of a bot that took the most effort to build, the Flows, Apex classes and integrations behind it, are exactly the parts the new agent can call as actions. The conversational layer is the part that gets redesigned, and it is also the part that benefits most from a fresh pass.
What is Salesforce Agentforce, and how is it different from a bot?
Salesforce Agentforce is Salesforce’s platform for building AI agents that work inside the CRM. Rather than following a scripted dialog tree, an Agentforce agent uses a large language model to understand what a customer is asking, decides which of its permitted actions will resolve it, and carries that action out against live Salesforce data. Those actions are usually Flows, Apex or API calls the organisation already trusts. Agents are grouped into topics with plain-language instructions, governed by the Einstein Trust Layer, and can hand over to a human through Omni-Channel when a conversation needs one.
An Einstein Bot, by contrast, matches a customer’s words to a trained intent and then walks a dialog the team designed in advance. That approach is predictable and has served service teams well for years. Agentforce keeps that predictability where it matters, through defined actions and instructions, while handling the long tail of phrasing and follow-up questions that a dialog tree struggles with. For customers, the difference shows up as fewer dead ends and fewer requests to rephrase.
Your bot’s history is the best design brief you own
The most valuable input to an upgraded agent is not the bot’s configuration, it is the record of what customers actually said to it. Bot transcripts show the questions people really ask, the words they use, where conversations stalled and which requests ended in a transfer. Read together, they form a ranked list of what the new agent should handle first. They also reveal which escalations were genuinely necessary and which happened only because the dialog tree had no branch for an obvious follow-up question.
This is where CRM data quality becomes a direct advantage. An agent can only act well on records and knowledge it can trust, so accurate case data, current knowledge articles and consistent contact records translate straight into better answers and fewer handovers. Clean data matters because every automated decision inherits it, and an agent makes far more decisions per day than a person does. Teams that spend a short sprint refreshing their top fifty knowledge articles before go-live typically see that effort repaid in the first month of resolution figures.
How long does an Einstein Bot upgrade take?
A simple Einstein Bot with a handful of intents and clean integrations can move to a reviewed, production-ready Agentforce agent in a matter of weeks, because the tool does the scaffolding and the Flows behind it carry across. A complex bot with dozens of dialogs, custom variables and detailed routing is a larger piece of work. Implementation partners who have published plans for this conversion describe four to six months end to end, split across an audit of two to three weeks, design of three to four weeks, a build of six to twelve weeks and a parallel run before cutover.
The parallel run is the part worth protecting, and the tool makes it easy, because the old bot keeps working throughout. It is also worth confirming the prerequisites early: Data Cloud (now Data 360) activated, the Einstein Trust Layer configured and Omni-Channel set up with enhanced messaging. None of these is a large task on its own, and each one also pays off for every other agent the organisation builds later, so the effort compounds well beyond this single project.
How do you prove the new agent is better?
The cleanest way to measure the return on a CRM investment is to fix a baseline before anything changes, pick a small number of outcome measures, and compare like with like over a defined period. For a service agent, the useful measures are resolution rate without a human, average handling time on transferred conversations, customer satisfaction on resolved chats and cost per resolved contact. Capture all four for the Einstein Bot now, while it is still handling full traffic, so the comparison with the new agent is fair and quick to read.
Salesforce shipped several tools at Dreamforce 2026 that make this measurement far easier than it used to be. Agent Health Monitoring is generally available with more than 16 built-in metrics covering error rates and latency. Voice and Conversation Testing, in beta, scores multi-turn conversations on a Task Resolution metric, and A/B Experimentation in Agentforce Builder enters beta in October. Together they let a team route a share of traffic to the new agent, compare it against the bot on the same questions and expand only when the numbers support it.
Bring the service team in from the first week
The people who handle escalations from the bot every day know its weak spots better than any report. Inviting two or three experienced service agents into the design phase, to review transcripts and test early versions of the new agent, is one of the most reliable ways to improve CRM adoption rates. Adoption rises when people can see their own knowledge shaping the tool, when the tool removes work they dislike, and when they have a clear role in improving it. A service agent that takes on repetitive password resets and order-status checks tends to win over a team quickly.
There is a useful bonus for European service teams. Agentforce Voice added Spanish, Italian, German, Portuguese, Japanese and Canadian French as generally available languages at Dreamforce 2026, which widens what the upgraded agent can cover across markets where an Einstein Bot may have been English-only. For organisations serving customers in Stockholm, Barcelona, Turin and beyond, that turns the upgrade from a like-for-like replacement into a genuine extension of service hours and reach. Treat it as a second release, once the core agent has proved itself in its first market.
The Sirocco perspective
We like this release because it rewards the work service teams have already done. An Einstein Bot that has run for three years is not legacy to be thrown away, it is a detailed, real-world specification of what customers need, plus a set of tested Flows and integrations ready to become agent actions. The Upgrade Agent tool gets a team to a first draft quickly, and the parallel run lets them earn confidence in the new agent on live conversations rather than on promises.
As an independent CRM partner working across Salesforce, HubSpot and Microsoft Dynamics 365, we see the same move from scripted chatbots to reasoning agents on every platform, and the teams that do well treat their conversation history and knowledge base as the real asset. Our role is to help clients get from the scaffolded agent to a measured, trusted production release, with routing, governance and adoption handled properly. If you have an Einstein Bot in place today, you already hold most of what a strong first Agentforce release needs, and now is a good moment to start. Schedule a consultation and we can map out your first release together.
Get in Touch
If you are weighing when and how to move your Einstein Bot onto Agentforce, we would be glad to look at your transcripts, flows and routing with you and map out a sensible first release.
