Every Major Salesforce AI Product from the Last Three Years
From Einstein GPT to Agentforce — a rundown of the AI products Salesforce has shipped since 2023 and what they actually do.
Salesforce has released more AI products in the last three years than in the previous decade combined. For teams trying to keep up, it can be hard to separate the meaningful launches from the rebrandings. Here’s a clear timeline of what shipped, what it does, and where it fits.
2023: The generative AI pivot
Einstein GPT (March 2023)
Salesforce’s first generative AI play. Einstein GPT brought large language model capabilities into Sales Cloud, Service Cloud, and Marketing Cloud — generating email drafts, summarizing case histories, and auto-creating knowledge articles. It connected to OpenAI’s models initially, with Salesforce positioning it as an open-architecture approach to generative AI in the CRM.
Einstein Copilot — early previews (late 2023)
Toward the end of 2023 Salesforce began previewing Einstein Copilot, a conversational AI assistant embedded directly into the Salesforce UI. Unlike Einstein GPT’s feature-by-feature integrations, Copilot was designed as a single interface where users could ask questions, trigger actions, and get contextual recommendations across any Salesforce cloud.
Data Cloud expansion
Data Cloud (formerly Genie) matured significantly in 2023, becoming the connective tissue that made AI features useful. By unifying customer data across systems in real time, it gave the AI models the context they needed to generate relevant outputs rather than generic ones.
2024: Agents take center stage
Einstein Copilot GA (February 2024)
Einstein Copilot went generally available with built-in actions for sales, service, and commerce workflows. Users could ask it to “summarize this opportunity” or “draft a close plan,” and it would pull from CRM data, execute multi-step reasoning, and return structured answers. Salesforce also introduced the Copilot Builder, allowing teams to create custom actions and prompt templates.
Agentforce (September 2024)
The biggest shift. At Dreamforce 2024 Salesforce introduced Agentforce — a platform for building and deploying autonomous AI agents that operate within Salesforce. Unlike copilots that assist a human, Agentforce agents can independently handle tasks: qualifying leads, resolving service cases, managing campaigns, and more. They run on the Atlas Reasoning Engine and are constrained by guardrails, topics, and defined actions.
Agentforce shipped with prebuilt agents for Service, Sales Development, and Commerce, plus a low-code Agent Builder for custom agents.
Agentforce was the moment Salesforce stopped selling AI features and started selling an AI platform. Everything since has been building on that foundation.
Prompt Builder and Einstein Trust Layer
Two supporting products rounded out the 2024 stack. Prompt Builder gave admins a visual tool for creating grounded prompt templates that pull in CRM data — without writing code. The Einstein Trust Layer became the security backbone, handling data masking, toxicity detection, and audit logging for all AI interactions.
2025: Deeper, more vertical
Agentforce 2.0 (early 2025)
Salesforce expanded Agentforce with a library of prebuilt skills, deeper MuleSoft integration for connecting to external systems, and the ability for agents to collaborate with each other on complex workflows. The agent-to-agent orchestration capability let teams build systems where a sales agent hands off to a service agent seamlessly.
Industry-specific AI
In 2025 Salesforce rolled out vertical AI solutions: Financial Services Cloud got AI-powered compliance monitoring and client summary generation. Health Cloud introduced clinical trial matching and patient communication agents. Manufacturing Cloud added predictive supply chain features. These weren’t generic models — they were fine-tuned on industry-specific data patterns.
Einstein for Developers
AI-assisted development went GA, with code generation, test creation, and Apex debugging built into the Salesforce development workflow. This brought Salesforce’s AI capabilities to the platform engineering layer, not just the business user layer.
What this means for teams
Three takeaways from this wave:
The platform is the product now. Salesforce is no longer selling point AI features — it’s selling an AI development platform. Agentforce, Data Cloud, and Prompt Builder are building blocks, not finished solutions. Teams that treat them as plug-and-play will underinvest in the architecture needed to make them work.
Data quality is the bottleneck. Every one of these products depends on clean, unified data. Data Cloud helps, but it doesn’t fix upstream problems. Organizations without a data strategy will get mediocre results regardless of how good the AI models are.
The skills gap is real. Building effective agents requires understanding both Salesforce configuration and AI reasoning patterns. It’s a new discipline — not a natural extension of Salesforce admin work and not a pure data science problem either. The teams that invest in cross-functional AI + Salesforce expertise will move faster than those staffing these roles separately.
The teams that win here won’t be the ones with the best AI models — they’ll be the ones with the cleanest data and the most cross-functional expertise.
The pace isn’t slowing down. Salesforce has signaled that AI agents will become the primary way users interact with the platform over the next few years. Whether that’s aspirational or inevitable depends on how well organizations build the foundations now.