For years, "Salesforce AI" meant Einstein. If you evaluated a Salesforce chatbot before 2024, you were looking at Einstein Bots and a set of Einstein-branded predictions bolted onto the CRM.
That's not the story anymore. Salesforce has reorganized almost its entire product line around a new brand, Agentforce, and the old Einstein names are being retired or absorbed. If you're evaluating Salesforce for AI customer service in 2026, the guide you read two years ago is describing a product that barely exists now.
This is what Salesforce AI actually is today: what Agentforce does, how it differs from the Einstein Bots you may remember, what it costs, and where a CRM-agnostic platform makes more sense.
What Is Salesforce AI in 2026?
Salesforce AI now comes in two layers, and the newer one is where all the momentum is.
The older layer is Einstein, the brand Salesforce has used since 2016 for its machine-learning and generative features. The customer-facing piece was Einstein Bots, a rule-based chatbot you built inside Service Cloud to greet customers, answer common questions, and route conversations to a human. It works, but it's a decision-tree bot: it handles one query at a time, follows the paths you scripted, and can't reason or act on its own.
The newer layer is Agentforce, launched in late 2024 and now Salesforce's flagship AI product. Agentforce is a platform for building autonomous AI agents that use large language models to interpret a request, reason about it, and take action across your Salesforce data, not just reply from a script. The customer-service version is the Agentforce Service Agent, the direct successor to Einstein Bots.
The practical takeaway: if someone says "the Salesforce AI chatbot" in 2026, they usually mean Agentforce. Einstein Bots still exists and is still supported, but Salesforce's new investment is going almost entirely into Agentforce. If you're starting fresh, that's the product to evaluate.
Salesforce Chatbot: Options at a Glance
A "Salesforce chatbot" can mean three different products with three different cost structures. Before you evaluate any single vendor, it helps to see the options side by side.
The first two options keep everything inside Salesforce, for better or worse. The third is how teams that connect agents to Salesforce by API keep their CRM while running the agent on their own terms. The rest of this guide covers all three in more depth, starting with the gap between Einstein Bots and Agentforce.
Agentforce vs. Einstein Bots: What Actually Changed
The gap between the two is the same gap that separates a scripted chatbot from a real AI agent.
Einstein Bots
Einstein Bots is rule-based. You map intents, write the responses, and maintain the tree. It answers one thing at a time and stays inside the paths you built. It's a reasonable choice for high-volume, predictable questions if you want something simple to govern.
Agentforce
Agentforce is agentic. It reasons over your CRM data, handles multi-step requests end to end, and can execute actions (update a case, book an appointment, process a return) rather than just suggesting them. This is the shift from agentic AI that the whole customer-service automation category is moving toward.
If you already run Einstein Bots, Salesforce's own guidance is to plan your roadmap around Agentforce. The rule-based bot is the past; the reasoning agent is where the product is going.
The Dreamforce 2025 Rebrand You Need to Know About
Here's the part that trips up anyone reading older documentation. At Dreamforce 2025, Salesforce renamed much of its stack around the Agentforce brand:
- Service Cloud became Agentforce Service.
- Data Cloud became Data 360.
- The Einstein 1 Platform became the Agentforce 360 Platform.
- Einstein Copilot, the assistant that helps your human reps inside the console, became the Service Assistant.
None of this changes what the products do, but it does mean the terminology in most "Salesforce Einstein" guides is now a version or two behind. When you're comparing notes with a vendor or a consultant, make sure you're both talking about the current names.
Salesforce Chatbot Setup and Channel Coverage
Setting up the Agentforce Service Agent starts inside Service Cloud. You connect it to your existing knowledge articles, case data, and flows. Then you define the actions it's allowed to take: look up an order, update a case, escalate to a human.
Einstein Bots setup is similar in shape, but simpler. You build an intent tree and script each branch by hand, with no reasoning layer to configure.
Channel coverage is where the two diverge most. Agentforce covers web chat, Slack, and, in the Service edition, phone through Salesforce's voice stack. Einstein Bots, by contrast, is scoped to web and messaging channels. Salesforce chatbots don't automatically cover voice: check which edition and add-ons actually include phone support before you budget for it. It's easy to assume a channel is covered because the brand exists, then discover it's a separate SKU.
The limit most teams hit is reach beyond Salesforce's own data model. The agent reasons well over what's in your CRM. Connecting it to a ticketing system, a separate knowledge base, or a non-Salesforce channel takes custom integration work, not a setup toggle.
How Much Does Salesforce AI Cost in 2026?
This is the question the older guides skip, and it's the one that matters most for a buying decision. Salesforce has changed Agentforce pricing more than once, and in 2026 there are three overlapping models you'll run into:
Per Conversation
Roughly $2 per conversation, a flat rate Salesforce positions for customer-facing agents like the Agentforce Service Agent. Salesforce calls Flex Credits the more flexible option, and you can't run both models in the same org. Check Salesforce's own pricing page for current rates before you commit.
Flex Credits
A consumption currency that covers agent actions, prompts, translations, and voice actions. It runs about $500 per 100,000 credits, where a standard action is around $0.10 and a voice action a bit more. Your bill scales with how much your agent actually does.
Per-User Add-Ons
Salesforce sells per-user licensing for employee-facing agents, with unmetered Agentforce use for licensed users. Add-ons for Sales, Service, and Field Service start around $125 per user per month, or $150 for Industries clouds. Max Edition, which includes Agentforce, starts at $550 per user per month.
The honest framing: Agentforce cost is consumption-based and stacks on top of the Salesforce licenses you already pay for. The "AI" line item is separate from your seats, and it grows with usage. Before you commit, model the real workload, not the demo. This is the same ROI math any serious enterprise AI rollout needs.
Is Salesforce AI Right for You?
It depends almost entirely on how committed you already are to Salesforce.
- Worth it if you're all-in on the Salesforce ecosystem. Agentforce reasons over the data already sitting in your CRM, and that native access to your accounts, cases, and history is the real advantage. If your team lives in Salesforce, keeping the agent there is a reasonable call.
- Harder to justify if you're not. The value is tied to the CRM and the ecosystem. If you're buying Salesforce mainly to get the AI agent, you're paying for a lot of platform to reach one feature, on a model you don't control, with a consumption bill that's hard to predict.
That trade, ecosystem depth versus flexibility and cost, is the decision. It's worth weighing against a dedicated agent platform before you commit.
The Best Salesforce AI Alternative
If you want an AI agent that isn't tied to one CRM's roadmap, model, or pricing meter, Voiceflow is the platform most teams compare against Agentforce. The honest distinction: Agentforce is an AI layer inside the Salesforce ecosystem, while Voiceflow is a dedicated platform for building and deploying agents into whatever stack you already run.
Where Voiceflow pulls ahead for serious agent builds:
- CRM-agnostic. Connect to Salesforce, Zendesk, HubSpot, or your own systems by API. You're not buying into one CRM's roadmap to get an agent.
- Model-agnostic. Run OpenAI, Anthropic, Google, or your own model, and switch as cost and quality change, instead of being tied to one vendor's model layer. If you're still comparing options, our roundup of the best AI chatbots and the primer on large language models go wider.
- Real agent primitives. Workflows for the deterministic steps that must run the same way every time, and Playbooks for open-ended reasoning, plus a built-in Knowledge Base so answers come from your content.
- Chat and voice. The same agent handles website chat, messaging, and phone, with a clean human handoff when it hits its limit.
- Built for production. Observability for turn-by-turn transcripts, Evaluations to score conversations against your own criteria, Environments to test in staging, and SOC 2 Type 2 with PII masking. Teams like Turo, StubHub International, Sanlam Studios, and Trilogy build their agents this way.
You don't have to choose one or the other. A common setup is to build the agent in Voiceflow and sync conversations and contacts back to Salesforce by API, so you keep the CRM you rely on and run an agent that isn't capped by it.
