The Three Waves of Salesforce AI: Why Your Einstein/AI Strategy Needs an Update

Justin Kasbohm – December 9, 2025

What Are the Three Waves of Salesforce AI and When Should You Use Each One?

If your AI strategy still centers around predictive scoring and dashboard insights, we have news: you're stuck in Wave One of a three-wave revolution. And while there's nothing wrong with predictive AI, which is still incredibly valuable; the game has fundamentally changed.

Salesforce has positioned Agentforce as the "third wave" of AI, building on predictive analytics and generative capabilities to deliver autonomous agents that can actually do things without constant human oversight. But here's what nobody's explaining clearly: each wave serves a different purpose or AI strategy, and you don’t need a particular one. You probably need all three.

The question isn't which wave to ride; it's knowing when to use predictive AI, AI assistants, or autonomous agents. Get this wrong, and you'll either over-engineer simple problems or under-deliver on complex ones.

Let's break down what each wave actually does, where it fits, and how to build a strategy that leverages all three intelligently.

Wave 1: Predictive AI as the Foundation That Still Matters

What it does: Analyzes historical data to forecast future outcomes and recommend next-best actions.

Predictive AI was Salesforce's original Einstein play, launching in 2016. It's the kind of AI strategy that scores your leads, predicts which opportunities will close, forecasts revenue, and flags customers at risk of churn. It doesn't generate content or take autonomous action; it simply tells you what's likely to happen and what you should do about it.

Where it excels:

  • Lead and opportunity scoring
  • Sales forecasting based on pipeline health
  • Churn prediction and customer lifetime value analysis
  • Product recommendations for e-commerce
  • Identifying patterns in large datasets

Real-world example: A sales team using Einstein Lead Scoring can prioritize the 20% of leads most likely to convert, rather than wasting time on cold prospects. A service team can proactively reach out to high-churn-risk accounts before they leave. This isn't flashy, but it's effective. And it's still generating over one trillion predictions per week across Salesforce's customer base.

When to use Wave 1: When you need data-driven insights to inform human decision-making. Predictive AI is perfect for scenarios where humans make the final call but need better intelligence to guide them. Think: "Which leads should I call today?" or "Which customers need attention this quarter?"

The catch: Predictive AI requires clean historical data and clear outcome variables. If your Salesforce data is messy or you don't have enough historical examples of what "success" looks like, predictions will be unreliable.

Wave 2: Generative AI and AI Assistants as the Productivity Multiplier

What it does: Creates new content, automates repetitive communication, and assists humans with suggestions in real-time.

The second wave arrived with Einstein GPT and later evolved into Einstein Copilot (now rebranded as Agentforce Assistant). Powered by large language models, this layer doesn't just predict, it generates. It drafts emails, summarizes meeting notes, creates case summaries, writes product descriptions, and even helps developers write code.

Agents work alongside humans. They monitor what you're doing, offer contextual suggestions, and can execute tasks when you ask, but they don't operate independently. You're still in the driver's seat; the agent is your highly capable assistant.

Where it excels:

  • Drafting personalized sales emails based on CRM context
  • Auto-generating customer service responses grounded in company knowledge
  • Summarizing sales calls, meetings, or support cases
  • Creating marketing campaign briefs and content
  • Helping admins and developers write formulas, flows, or Apex code

Real-world impact: A customer service rep handling a complex case can ask the agent to summarize the entire case history, draft a response in line with company policies, and suggest next steps, all in seconds. A sales rep preparing for a call can have the sales AI agent analyze past customer interactions, assess sentiment, and draft a personalized follow-up email. The human reviews, refines, and sends.

When to use Wave 2: When humans are doing valuable work but spending too much time on mechanical or repetitive parts of that work. People shine in scenarios that require judgment, creativity, or relationship management, where agents can handle the grunt work but humans need to stay in control.

Research shows agents can improve employee productivity by 5-10%. That may not sound like much at first, but it's meaningful. Especially when applied across large teams handling high volumes of communication or documentation.

The catch: Agents are only as good as the data they're grounded in. If your Data 360, formerly Data Cloud (collective naming sigh) isn't implemented or your CRM data is incomplete, agentic suggestions will be generic or wrong. Also, agents require continuous human input; the training and data updates that they can't run autonomously, which means they don't scale for truly repetitive, rules-based work.

Wave 3: Autonomous AI Agents as the Workforce Extension

What it does: Operates independently to complete entire workflows from start to finish, reasoning through problems and taking action without human oversight at every step.

This is where Agentforce 3.0 comes into play. Autonomous agents don't just assist; they act. Powered by the Atlas Reasoning Engine, these agents can handle multi-step processes, make contextual decisions, and escalate to humans only when necessary.

Unlike chatbots that follow rigid decision trees, agents dynamically reason through problems. They retrieve relevant data, evaluate options, generate plans, and refine their approaches based on outcomes. They're designed for "System 2" reasoning: the kind of deliberate, conscious, multi-step thinking humans use to solve complex problems.

Where it excels:

  • 24/7 customer service for common inquiries (password resets, order status, basic troubleshooting)
  • Employee support (HR questions, IT helpdesk, onboarding workflows)
  • Sales development (qualifying leads, scheduling meetings, nurturing prospects)
  • Order management and returns processing
  • Appointment scheduling and reminders

Real-world impact: During tax season 2025, 1-800Accountant's Agentforce agent autonomously resolved 70% of administrative chat engagements, freeing CPAs to handle complex tax questions. Engine's service agent handles 30% of customer support cases autonomously, reducing average handle time by 15% and saving an estimated $2 million annually.

Studies show autonomous AI agents can boost efficiency by 20-50% for tasks they're designed to handle, dramatically higher than AI assistant-level improvements.

When to use Wave 3: When you have high-volume, repetitive workflows with clear business rules and escalation paths. Agents work best for tasks you can document clearly, where 70-80% of cases follow predictable patterns and the remaining cases can be escalated to humans.

The catch: Agents require significant upfront work. Your data must be clean and unified (ideally through Data Cloud). Your processes must be well-documented. You need to set clear guardrails and define escalation triggers. And you must be willing to iterate. The best Agentforce implementations treat agents like products, continuously testing and refining them. (ask Midwest Dreamin’ Demo Jam First Runner Up, Andrew Dawson, about testing and refining agents!)

Agents also struggle with nuanced judgment calls, complex multi-stakeholder workflows, and scenarios requiring empathy or reading between the lines. If your process requires deep human intuition, agents aren't ready yet.

The Decision Framework: Which AI Strategy When?

Here's a simple framework to help you choose:

Use Predictive AI (Wave 1) when:

  • You need data-driven insights to inform human decisions
  • You're trying to prioritize among many options (leads, opportunities, customers)
  • Success depends on recognizing patterns in historical data
  • Humans make the final call, but need better intelligence

Use AI Assistants (Wave 2) when:

  • Humans are doing valuable work, but spending too much time on repetitive subtasks
  • Tasks require judgment, creativity, or relationship management
  • You want to accelerate workflows without removing human oversight
  • Content generation or summarization would save significant time

Use Autonomous Agents (Wave 3) when:

  • You have high-volume, well-defined, repetitive tasks
  • 70-80% of cases follow predictable patterns
  • Speed and 24/7 availability matter more than nuanced judgment
  • You can clearly document the process and escalation rules
  • Your data infrastructure is solid (Data Cloud implemented, clean CRM data)

The Reality: You Need All Three (But Not Everywhere)

Here's the insight most organizations miss: the three waves aren't competing—they're complementary.

Consider a customer service scenario:

  • Predictive AI scores incoming cases by severity and churn risk, routing high-value or at-risk customers to your best agents
  • AI assistants assist humans on the remaining 30% of complex cases, summarizing case history, suggesting responses, and drafting detailed explanations
  • Autonomous agents handle 70% of routine inquiries (password resets, order status, basic troubleshooting) without additional human intervention

Or a sales scenario:

  • Predictive AI scores and ranks leads based on the likelihood to convert
  • AI Assistants help sales reps prepare for calls with high-priority leads, draft personalized follow-ups, and summarize conversations.
  • Autonomous agents qualify low-priority leads, schedule initial meetings, and nurture prospects through email sequences

The Magic Happens 🧙🏻‍♂️:
When you strategically layer these capabilities, you use the right tool for each part of your workflow.

Common AI Strategy Mistakes We're Seeing

1. Trying to use agents for everything. Not every problem needs autonomous AI. If a task requires frequent human judgment or your process isn't standardized, you're better off with agents assisting humans than agents trying (and failing) to operate independently.

2. Ignoring Wave 1 because it's "old." Predictive AI isn't sexy, but it's still the most reliable way to prioritize and forecast. Don't abandon lead scoring or opportunity insights just because generative AI is getting all the buzz.

3. Deploying AI assistants without proper data grounding. If your assistant isn't connected to Data Cloud or your CRM data is incomplete, it will generate generic, unhelpful suggestions. Data quality is non-negotiable.

4. Building agents without documenting processes first. If you can't clearly articulate what the agent should do in every scenario, you're not ready to build it. Start by documenting your workflows using simple concepts and words. If that's hard, pause and fix your processes before automating them.

What This Means for Your AI Strategy

If your organization is still treating "Einstein" as a single monolithic thing, it's time to rethink. You need a multi-layered AI strategy that deploys the right capabilities in the right places. Learn about the Strategic Growth Ally in a concept and blog by EVP, Client Success, Jessica Taylor.

Start with an audit:

  1. Where are you currently using predictive AI, and is it delivering value? (Lead scoring, forecasting, churn prediction)
  2. Where could AI assistants accelerate human work without requiring full automation? (Email drafting, case summarization, meeting prep)
  3. What high-volume, repetitive workflows could autonomous agents handle? (Routine service inquiries, lead qualification, appointment scheduling)

Then build intentionally:

  • Don't skip Wave 1 thinking it's outdated; predictive insights remain foundational
  • Deploy AI assistants where humans do valuable work but waste time on mechanical subtasks
  • Reserve autonomous agents for workflows you've mastered, documented, and are ready to scale

And remember: The best implementations don't pick one wave—they layer all three intelligently, letting each do what it does best.

The bottom line: Salesforce's three waves of AI aren't a progression where newer is always better. They're a toolkit, and the art is knowing which tool fits which job. Predictive AI informs decisions. Agents accelerate human work. Autonomous agents handle repetitive tasks at scale.

So, when are companies winning with AI? When they're not chasing the latest wave. When they're strategically deploying all three where they make sense, they're building a comprehensive AI capability that's greater than the sum of its parts.

Ready to audit your AI strategy and figure out where each wave fits in your organization? Let's talk. Digital Mass will help you build a roadmap that's grounded in what actually works, not just what sounds impressive on stage.

Further Reading