Skip to content
WINTER '27 STACK•AI-201 Aligned
Request Price, Duration & Next Batch Details
GetSalesforce.com
by GSF Talent Solutions
Back to All Engineering Guides
AI-201 CertificationStudy RoadmapData 360 RAGPrompt Engineering

Salesforce Certified Agentforce Specialist (AI-201) Study Roadmap: Complete 6-Domain Engineering Guide

A structured preparation blueprint for the Salesforce Certified Agentforce Specialist (AI-201) exam—covering all 6 official domains, exact exam weights, architectural trade-offs, and hands-on lab strategies.

GSF Talent Solutions Engineering7 min read

The Salesforce Certified Agentforce Specialist (AI-201) certification has rapidly become the benchmark credential for Salesforce Developers, Technical Architects, and Senior Administrators building autonomous AI solutions on the Winter '27 stack.

Unlike introductory AI exams that test high-level terminology, AI-201 evaluates your ability to make production architectural decisions across the Atlas Reasoning Engine, Agent Script guardrails, Prompt Builder grounding, Data 360 Hybrid RAG, Multi-Agent (A2A & MCP) interoperability, and the Einstein Trust Layer.

In this comprehensive study roadmap, we break down the 6 official AI-201 exam domains, their exact percentage weights, the most frequently tested architectural trade-offs, and a week-by-week hands-on lab strategy.


1. Official AI-201 Exam Blueprint & Domain Weights

Every study plan should be anchored to the official domain weightings. Notably, 75% of the entire exam is concentrated in the first three domains: AI Agents (35%), Prompt Engineering (20%), and Data 360 for AI (20%).

Domain #Official AI-201 Exam DomainExam WeightPrimary Winter '27 Stack Focus
Domain 1AI Agents35%Atlas Reasoning Engine, Topics, Instructions, Agent Script guardrails, Action contracts (Apex, Flow, External Services)
Domain 2Prompt Engineering20%Prompt Builder (Field Generation, Flex, Record Prioritization), dynamic grounding via Flow & Apex merge fields
Domain 3Data 360 for AI20%Zero-Copy federation, Unstructured Data Lakes, chunking strategies, Vector Database indexing, Hybrid RAG retrievers
Domain 4Development Lifecycle10%Agentforce DX (sf agent CLI), Agent Testing Center batch evaluations, Event Logs, Utterance Analysis
Domain 5Multi-Agent Interoperability10%Supervisor-Worker orchestration, Agent-to-Agent (A2A) protocols, Model Context Protocol (MCP) servers, HXL LWC surfaces
Domain 6Trust & Governance5%Einstein Trust Layer, Zero-Data Retention, dynamic PII masking, toxicity scoring, Prompt Defense guardrails

2. Deep Dive by Domain: What the Exam Actually Tests

Domain 1: AI Agents (35% Weight)

Domain 1 is the single largest section of the AI-201 exam. Expect scenario-based questions asking you to diagnose why an agent selected the wrong action or how to enforce strict compliance rules.

  • Topics vs. Actions: A Topic defines a category of job-to-be-done with classification descriptions and scoped instructions. Actions are the executable tools (@InvocableMethod Apex, Autolaunched Flows, Prompt Templates, or External Services) assigned to that Topic.
  • When to Use Agent Script vs. Natural-Language Instructions: Use natural-language Topic instructions to guide conversational tone and slot collection, but use deterministic Agent Script (when, invoke, else_escalate) whenever a business rule mandates a hard numeric threshold, precondition verification, or mandatory human escalation.
  • User Mode & Security Boundaries: Understand how agent actions execute under the context of the assigned agent user profile/permission set and why WITH USER_MODE in Apex is critical to prevent unauthorized record exposure.

Domain 2: Prompt Engineering (20% Weight)

Domain 2 tests your mastery of Prompt Builder and how structured CRM data is injected into LLM prompts securely and token-efficiently.

  • Choosing the Right Prompt Template Type:
    • Field Generation: Populates a single field on a Lightning Record Page (e.g., summarizing an escalated Case history).
    • Flex Template: Accepts arbitrary object inputs from Flows, Apex, or Agentforce actions to handle multi-object synthesis.
    • Sales Email / Record Prioritization: Specialized templates with pre-wired recipient and workspace contexts.
  • Grounding Mechanisms: Know when to use Record Merge Fields (direct parent/child fields), Flow Merge Fields (filtering related lists and pre-aggregating metrics without burning tokens), and Apex Merge Providers (complex cross-object transformations or external callouts).

Domain 3: Data 360 for AI (20% Weight)

An autonomous agent is only as accurate as its grounding layer. Domain 3 evaluates how you ingest, chunk, index, and retrieve enterprise knowledge inside Data 360 (formerly Data Cloud).

  • Zero-Copy Data Federation: Understand how Zero-Copy connectors query Snowflake, BigQuery, Redshift, and Databricks without duplicating raw tables into Salesforce storage.
  • Chunking Strategies for Unstructured Documents:
    • Passage / Semantic Chunking: Best for policy PDFs, legal contracts, and technical manuals where context spans paragraphs.
    • HTML / Header-Aware Splitters: Best for structured knowledge base articles with clear section hierarchy.
    • Conversation / Transcript Chunking: Preserves speaker turns in service call logs and Slack threads.
  • Hybrid RAG Search (Vector + Keyword BM25): Know why Hybrid Search outperforms pure vector similarity when users query exact alphanumeric identifiers (such as SKU numbers, error codes, or invoice IDs) alongside natural-language questions.
-- Conceptual Data 360 Hybrid Retriever Evaluation Query
SELECT
  chunk_id,
  document_title,
  hybrid_score(vector_embedding, keyword_index, 'ORD-88412 SLA credit policy') AS relevance_score
FROM Data360_DisputePolicy_Index
WHERE relevance_score >= 0.78
ORDER BY relevance_score DESC
LIMIT 5;

Domain 4: Development Lifecycle (10% Weight)

Domain 4 focuses on testing, observability, and deployment across sandboxes and production orgs:

  • Agent Testing Center: Evaluates batch CSV/YAML test cases against expected topic classification, action invocation, and response completeness metrics.
  • Agentforce DX CLI (sf agent): Automates metadata validation, scratch org synchronization, and CI/CD test execution.
  • Agent Event Logs & Utterance Analysis: Identifies low-confidence intent routing, unhandled customer questions, and latency bottlenecks in production conversations.

Domain 5: Multi-Agent Interoperability (10% Weight)

Enterprise architectures rarely rely on a single monolithic agent. Domain 5 covers:

  • Supervisor-Worker Topology: A central Supervisor Agent decomposes complex requests and delegates specialized tasks to domain-specific Worker Agents (e.g., Inventory Logistics Worker, Billing & Credit Worker, Compliance Auditor Worker).
  • Model Context Protocol (MCP) & Agent-to-Agent (A2A): Standardized contracts allowing Agentforce to discover external tools, query third-party enterprise systems, and hand off state payloads cleanly.
  • Human-in-the-Loop Experience Layer (HXL): Surfacing agent reasoning traces and one-click approval cards inside custom Lightning Web Components (LWC).

Domain 6: Trust & Governance (5% Weight)

Even at 5% exam weight, Einstein Trust Layer principles appear interwoven across scenarios in Domains 1, 2, and 3:

  • Request Pipeline: Dynamic Grounding $\rightarrow$ PII/PCI Data Masking (format-preserving tokens like [PERSON_1]) $\rightarrow$ Prompt Defense Guardrails $\rightarrow$ LLM Gateway.
  • Response Pipeline: Zero-Data Retention enforcement (external LLM providers never store or train on customer payloads) $\rightarrow$ Toxicity & Safety Scoring $\rightarrow$ Data Demasking $\rightarrow$ Audit Trail logging in Data 360.

3. Recommended 6-Week Hands-On Lab Preparation Strategy

Reading documentation alone is rarely enough to pass scenario questions that contrast subtle configuration options. We recommend pairing each study week with a concrete implementation lab:

  1. Week 1 (Foundations & Vibes IDE v4): Configure your Winter '27 developer environment, connect the Salesforce DX MCP server, and inspect Agent.yaml metadata bundles.
  2. Week 2 (Domain 1 + Domain 6 — Agent Script & Trust Layer): Build a deterministic customer service agent using Agent Script preconditions and verify PII token masking in the Einstein Trust Layer audit logs.
  3. Week 3 (Domain 2 — Prompt Builder & Invocable Apex): Author a Flex Prompt Template grounded by both a Flow Merge Field and a bulkified @InvocableMethod Apex class running in USER_MODE.
  4. Week 4 (Domain 3 — Data 360 Hybrid RAG): Ingest a set of enterprise SLA PDFs into Data 360, configure passage chunking with overlap, build a Hybrid Retriever, and bind it to an Agentforce action.
  5. Week 5 (Domain 5 — Multi-Agent Orchestration & HXL): Wire a Supervisor Agent to two Worker Agents and build an HXL Lightning Web Component that renders live confidence scores and escalation controls.
  6. Week 6 (Domain 4 — Agent Testing Center & Exam Simulation): Run a 50-utterance adversarial test suite via sf agent test run, diagnose utterance routing logs, and complete full-length timed AI-201 practice scenarios.

Key Takeaway

The Salesforce Certified Agentforce Specialist (AI-201) exam rewards engineers who understand how deterministic code, structured retrieval, and probabilistic reasoning work together. By mastering Agent Script, Data 360 Hybrid RAG, and the Agentforce DX testing lifecycle, you not only secure the certification—you gain the architectural toolkit required to lead enterprise Agentforce implementations.

THE AGENTIC SALESFORCE ENGINEER

Master This Architecture in Our Next Live Engineering Cohort

Get the complete 6-module Winter '27 syllabus, cohort fee structure, program duration, and upcoming batch schedule from GSF Talent Solutions.

  • 6 Core Modules + 6 Hands-On Labs
  • 100% AI-201 Exam Domain Alignment
  • 4-Pillar Enterprise Dispute Capstone
  • Agentforce DX, Agent Script & Data 360