The Agentic Salesforce Engineer — Complete Syllabus
Architect, Script, Ground, Test & Deploy Production Agentforce Systems on the Salesforce Winter ’27 Stack
Salesforce Developers (Apex & LWC)
Transition from imperative CRUD triggers and controllers to deterministic Agent Script, Invocable AI Actions, Headless Agent APIs, and HXL surfaces.
Salesforce Technical Architects & Tech Leads
Design enterprise-grade multi-agent topologies, Data 360 Zero-Copy RAG pipelines, Einstein Trust Layer governance, and Agentforce DX CI/CD lifecycles.
Senior & Code-Curious Salesforce Admins
Move beyond point-and-click basics into Prompt Builder engineering, Autolaunched Flow agent actions, Agent Testing Center batch evaluations, and AI-201 mastery.
AI & Full-Stack Engineers Entering Salesforce
Leverage Claude Code, Cursor Plugins, and Salesforce DX MCP Servers to build governed enterprise agents integrated with CRM and ERP systems.
- Working familiarity with core Salesforce administration (Objects, Fields, Permissions, Flows) or development (Apex, SOQL, LWC).
- Comfort reading declarative configurations, JSON payloads, or structured scripting syntax.
- No prior Machine Learning or Python training required — all AI engineering is taught from first principles on the Salesforce Winter ’27 stack.
Showing 6 Core Modules, 24 Technical Lessons & 6 Guided Hands-On Labs
Module 01 Engineering Lessons
From Imperative Automation to Autonomous Enterprise Agents
Contrast traditional Apex triggers, batch jobs, and rigid screen flows with goal-driven agentic loops powered by the Atlas Reasoning Engine in Winter ’27.
- Deterministic vs. probabilistic execution boundaries
- Atlas Reasoning Engine loop: Plan, Evaluate, Ground, Act
- When to use Flow/Apex vs. Single Agent vs. Multi-Agent systems
Agentforce Vibes IDE v4, Vibe Codey & Parallel Sub-Agents
Configure Agentforce Vibes IDE v4 and orchestrate Vibe Codey alongside Parallel Sub-Agents to generate metadata-aware Apex, LWC, and test classes simultaneously.
- Workspace indexing and Org schema grounding in Vibes IDE v4
- Delegating multi-file refactors to Parallel Sub-Agents
- Prompting Vibe Codey for governor-safe Invocable Apex actions
Salesforce Hosted & DX MCP Servers + Claude Code & Cursor Plugins
Wire Salesforce Hosted MCP and local DX MCP Servers into Claude Code and Cursor so external coding agents can query org metadata, execute SOQL, and validate deployments safely.
- Model Context Protocol (MCP) architecture for Salesforce DX
- Configuring Salesforce Hosted MCP & local `sf mcp` tool servers
- Integrating Claude Code & Cursor Plugins with scratch orgs and sandboxes
Einstein Trust Layer, Zero-Data Retention & Code Analyzer Gateways
Establish enterprise security guardrails from day one: PII masking, toxicity detection, zero-data retention policies, and automated static checks with Code Analyzer.
- Einstein Trust Layer request/response lifecycle & dynamic masking
- Zero-Data Retention (ZDR) and audit trail logging in Data 360
- Enforcing Apex and Agent action quality gates via Salesforce Code Analyzer
Hands-On Lab 1: Configuring the Winter ’27 Agentic Workstation & MCP Pipeline
Set up Agentforce Vibes IDE v4, connect Salesforce DX MCP Servers to Claude Code/Cursor, orchestrate Parallel Sub-Agents to scaffold an Invocable Action suite, and pass all Code Analyzer security checks.
Lab Deliverables
- Configured `.mcp.json` connecting Salesforce DX MCP Server to local IDE
- Scaffolded Apex service + test suite generated via Parallel Sub-Agents
- Zero-violation Code Analyzer static analysis report
Winter '27 Tools Used
Module 02 Engineering Lessons
Deep Dive into the Atlas Reasoning Engine & Topic Classification
Understand how Atlas evaluates user utterances, classifies intent into Topics, maintains conversational memory, and selects optimal action sequences.
- Topic classification descriptions vs. execution instructions
- Dynamic planner reasoning steps and latency optimization
- Preventing topic collision and ambiguous routing across large agents
Authoring Deterministic Specifications in Agent Script
Write structured `.agent` files in Agent Script to enforce deterministic state machines, explicit variable bindings, and predictable multi-step execution paths.
- Agent Script syntax: blocks, variables, conditions, and transitions
- Combining deterministic state rules with LLM reasoning flexibility
- Source-controlling `.agent` definitions in Git with Agentforce DX
System Instructions, Persona Engineering & Guardrail Policies
Craft unambiguous Topic instructions and defense-in-depth guardrails that prevent prompt injection, unauthorized commitments, and out-of-scope hallucinations.
- Positive constraint framing and deterministic fallback rules
- Input/output guardrail policies and escalation triggers
- Handling sensitive commercial disputes and compliance boundaries
Standard vs. Custom Agent Types & Security Contexts
Configure Service Agents, Sales Development (SDR) Agents, Employee Agents, and custom headless agents while respecting Object-Level and Field-Level Security (OLS/FLS).
- Agent user contexts, Permission Sets, and Einsteintrust execution rights
- Employee Agent vs. Customer-facing Service Agent security boundaries
- AI-201 blueprint scenarios on agent licensing and governance
Hands-On Lab 2: Authoring a Deterministic Enterprise Agent in Agent Script
Write, compile, and validate a multi-topic `.agent` specification using Agent Script that enforces strict verification preconditions before allowing order modification actions.
Lab Deliverables
- Source-controlled `.agent` specification with 4 deterministic topics
- State-transition guardrails blocking unverified account mutations
- Interactive CLI trace verification using `sf agent preview`
Winter '27 Tools Used
Module 03 Engineering Lessons
Data 360 Fundamentals, Data Streams & Zero-Copy Federation
Architect Data 360 ingestion pipelines, map Data Lake Objects (DLOs) to Data Model Objects (DMOs), unify identity resolution rules, and federate external lakes via Zero-Copy.
- Data Streams, DLOs, DMOs, and Unified Individual identity graphs
- Zero-Copy federation with Snowflake, Databricks, and BigQuery
- Calculated Insights and real-time data graphs for agent context
Document AI, Chunking Strategies & Vector Search Indexing
Ingest unstructured PDFs, commercial invoices, and policy manuals through Document AI, configure optimal chunking strategies, and build Vector Search indexes.
- Document AI schema extraction and confidence thresholds
- Passage chunking, overlap tuning, and vector embedding models
- Search Index configuration in Data 360 for low-latency retrieval
Hybrid RAG Retrievers (Keyword + Vector) & Custom Filters
Combine dense vector similarity with lexical keyword matching (BM25/Hybrid Search) and bind dynamic metadata pre-filters to eliminate cross-tenant or stale citations.
- Default vs. Custom Data 360 Retrievers in Einstein Studio
- Hybrid Search ranking weights, top-K tuning, and recency boosting
- Binding runtime record context into Retriever filter conditions
Prompt Builder Engineering: Flex, Field Generation & Record Summary Templates
Design reusable, grounded Prompt Builder templates using Merge Fields, Flow Data Providers, Apex Data Providers, and Data 360 Search Retrievers.
- Selecting Field Generation vs. Record Summary vs. Flex Templates
- Grounding prompts with Flow and `@InvocableMethod` Apex providers
- Resolution inspection, token budgeting, and AI-201 Prompt Builder scenarios
Hands-On Lab 3: Building a Data 360 Hybrid RAG & Document AI Grounding Pipeline
Parse enterprise dispute PDFs with Document AI, index policy documents in Data 360 Vector Search, configure a filtered Hybrid Retriever, and wire it into a Flex Prompt Template.
Lab Deliverables
- Document AI extraction configuration for dispute claim invoices
- Data 360 Hybrid Search Index + Custom Retriever with status filters
- Production Flex Prompt Template grounded in Data 360 + Apex context
Winter '27 Tools Used
Module 04 Engineering Lessons
Designing High-Precision Custom Agent Actions & I/O Contracts
Write action descriptions, input parameter constraints, and output schemas that the Atlas Reasoning Engine can reliably select and chain together.
- Action instruction semantics: how Atlas chooses between 15+ actions
- Typed input/output schemas, required flags, and confirmation prompts
- Chaining Prompt Template Actions with transactional Flow/Apex Actions
Pro-Code Agent Actions with Invocable Apex & Governor Safety
Build bulkified `@InvocableMethod` Apex classes with complex inner `@InvocableVariable` DTOs, custom exception handling, and structured error feedback for agent self-correction.
- Designing nested Apex request/response wrappers for Agentforce
- Graceful error serialization so the agent can recover and retry safely
- Enforcing `with sharing`, stripInaccessible, and User Mode DML
Autolaunched Flows, External Services & MuleSoft API Orchestration
Expose OpenAPI 3.0 specifications, External Services, Named Credentials, and MuleSoft APIs as governed Agentforce actions for real-time ERP and payment gateway operations.
- Wrapping External Services and HTTP Callouts inside Autolaunched Flows
- Named Credentials, OAuth 2.0 scopes, and idempotency keys
- Handling synchronous vs. asynchronous callout timeouts in agent turns
Human-in-the-Loop Experience Layer (HXL) & Custom Lightning Types
Render interactive Lightning Web Components (LWC) inside the agent conversation and enforce human approval gates for high-value financial or compliance actions.
- Architecture of HXL (Human-in-the-Loop Experience Layer)
- Creating Custom Lightning Types and LWC action renderers
- Human approval handoffs before executing high-value refunds or write-offs
Hands-On Lab 4: Transactional Apex/ERP Agent Actions with HXL Approval Cards
Build an Invocable Apex + External Service action suite that queries live ERP shipment telemetry, calculates refund eligibility, and renders an interactive HXL LWC approval card.
Lab Deliverables
- Bulkified `@InvocableMethod` Apex class with typed DTO inputs/outputs
- External Service callout wired via Named Credentials
- Custom HXL Lightning Web Component rendering structured refund previews
Winter '27 Tools Used
Module 05 Engineering Lessons
Multi-Agent Architecture: Supervisor, Router & Specialist Worker Agents
Design modular multi-agent topologies where a Triage/Supervisor Agent delegates specialized sub-tasks to domain-focused Worker Agents with isolated contexts.
- Single-agent context bloat vs. multi-agent domain isolation
- Supervisor-to-Worker delegation contracts and shared session state
- Preventing infinite agent-to-agent delegation loops
Agent-to-Agent (A2A) Protocols & Cross-System Orchestration
Coordinate Salesforce Agentforce agents with external enterprise agents and MCP tool servers using structured handoff payloads and deterministic routing.
- Agent-to-Agent (A2A) message envelopes and context preservation
- Combining Agent Script state rules with multi-agent handoffs
- Security and auditability across multi-agent boundaries
Headless Agent API: Event-Driven & Webhook Invocations
Invoke Agentforce programmatically from Apex, Platform Events, external microservices, or backend webhooks using the Headless Agent REST and Streaming APIs.
- Headless Agent API authentication, session lifecycle, and streaming
- Triggering autonomous background agents from Platform Events & CDC
- Parsing structured JSON responses from headless agent executions
Omni-Channel Deployment: Experience Cloud, Enhanced Messaging & Slack
Deploy customer and employee agents across Experience Cloud portals, WhatsApp/Enhanced Web Messaging, and Slack workspaces with seamless human rep escalation.
- Omni-Channel Flow routing and pre-chat context injection
- Deploying Agentforce in Slack with rich block-kit & HXL actions
- Warm transfer with structured conversation summaries to human reps
Hands-On Lab 5: Orchestrating a Multi-Agent Network & Headless API Webhook
Configure a Supervisor Triage Agent that delegates order logistics to a Fulfillment Worker Agent and financial claims to a Dispute Worker Agent, invokable via both Experience Cloud and the Headless Agent API.
Lab Deliverables
- Multi-agent delegation configuration linking Supervisor and 2 Worker Agents
- Headless Agent API integration script triggered by inbound order events
- Omni-Channel escalation flow with structured handoff summary
Winter '27 Tools Used
Module 06 Engineering Lessons
Agentforce DX (`sf agent`) & Source-Driven Metadata CI/CD
Manage the complete metadata lifecycle for GenAiPlannerBundle, GenAiPlugin, GenAiFunction, Prompt Templates, and `.agent` specs across scratch orgs, sandboxes, and production.
- Core metadata types: `GenAiPlannerBundle`, `GenAiPlugin`, `GenAiFunction`, `GenAiPromptTemplate`
- CLI workflows: `sf agent generate`, `sf agent preview`, `sf project deploy start`
- GitHub Actions CI/CD pipelines for automated agent validation
Automated Evaluation at Scale with Agent Testing Center
Generate synthetic and adversarial evaluation datasets (100+ utterances) and run batch tests in Agent Testing Center to score topic classification, action selection, and outcome quality.
- Authoring CSV/JSON test suites for Agent Testing Center
- Running `sf agent test run` in CI/CD and setting pass thresholds
- Diagnosing regressions when changing system instructions or models
Observability, Data 360 Audit Analytics & Agent Optimizer
Inspect step-by-step reasoning traces, latency bottlenecks, token consumption, and guardrail triggers using Agent Optimizer and Einstein Audit & Feedback data models.
- Analyzing session traces and reasoning steps in Agent Optimizer
- Querying Einstein Generative AI Audit & Feedback DMOs in Data 360
- Iterative prompt and action description tuning based on production telemetry
Salesforce Certified Agentforce Specialist (AI-201) Exam Fast-Track
Synthesize all 6 official AI-201 exam domains through architectural decision trees, distractor elimination strategies, and scenario-based mock walkthroughs.
- Complete review of all 6 AI-201 domains (35% / 20% / 20% / 10% / 10% / 5%)
- High-frequency scenario traps: Retriever filters, Permission Sets, and Action vs. Prompt selection
- Exam-day time management and architecture checklist
Hands-On Lab 6: CI/CD Batch Evaluation in Agent Testing Center & Trace Optimization
Execute a 50-scenario automated evaluation suite in Agent Testing Center via CLI, analyze failing edge-case traces in Agent Optimizer, refine the Agent Script & action descriptions, and achieve >96% accuracy.
Lab Deliverables
- 50-utterance adversarial test dataset for Agent Testing Center
- CLI batch test report (`sf agent test run`) with before/after accuracy scores
- Agent Optimizer trace diagnostic report and metadata deployment package
Winter '27 Tools Used
Autonomous Enterprise Order & Dispute Resolution System
4-Pillar Production Capstone Project • Built on the Winter ’27 Agentforce & Data 360 Stack
Architect and ship an end-to-end enterprise dispute, shipment claim, and order modification platform. Your system ingests unstructured customer claim PDFs, grounds reasoning in Zero-Copy ERP order telemetry via Data 360, coordinates a Supervisor + Specialist Worker multi-agent network via Agent Script, executes governed Apex/Flow refunds with HXL human approval thresholds, and passes automated CI/CD evaluation gates.
Pillar 1: Unstructured Evidence Ingestion & Data 360 Hybrid Grounding
Data 360 Zero-Copy • Document AI • Hybrid Vector + Keyword RAG
Ingest commercial invoices, carrier damage proofs, and SLA policy PDFs via Document AI while federating live ERP order lines through Data 360 Zero-Copy. Expose a filtered Hybrid Retriever to ground every dispute evaluation in verifiable facts.
- Document AI schema extraction pipeline for claim attachments
- Data 360 DLO-to-DMO mapping with Unified Account & Order identity
- Custom Hybrid Search Retriever filtered by contract tier and region
Pillar 2: Deterministic Agent Script & Multi-Agent Orchestration
Agent Script (.agent) • Supervisor & Specialist Worker Agents
Author a deterministic Triage Supervisor Agent in Agent Script that classifies incoming claims and coordinates two specialist Worker Agents: an Order Logistics Verifier and a Financial Dispute Resolver.
- Source-controlled `.agent` specification enforcing strict state transitions
- Supervisor-to-Worker A2A delegation with isolated topic instructions
- Einstein Trust Layer PII masking and zero-hallucination guardrail policies
Pillar 3: Transactional Action Execution, Headless API & HXL Surfaces
Invocable Apex • Autolaunched Flows • Headless Agent API • HXL LWC
Execute autonomous credit memos and replacement shipments under policy thresholds via Invocable Apex and External ERP APIs, while surfacing interactive HXL Lightning Web Component approval cards for high-value exceptions.
- Bulkified `@InvocableMethod` Apex actions with External Service callouts
- Headless Agent API webhook endpoint for automated carrier claim events
- HXL Custom Lightning Type rendering side-by-side dispute evidence & approval controls
Pillar 4: Agentforce DX CI/CD, Testing Center & Observability
Agentforce DX (`sf agent`) • Agent Testing Center • Agent Optimizer
Package the entire multi-agent solution as source-driven metadata, validate it against adversarial test cases in Agent Testing Center, and tune latency and token utilization with Agent Optimizer.
- Complete `sfdx-project.json` package with Planner, Plugins, Functions & Prompts
- Automated Agent Testing Center evaluation suite scoring routing & groundedness
- Agent Optimizer telemetry audit &Code Analyzer security sign-off
Graduating Portfolio Deliverables
Salesforce Certified Agentforce Specialist (AI-201) Alignment Table
Every official AI-201 exam domain and weighting is systematically covered across Modules 01–06 and the Hands-On Labs.
| Official AI-201 Domain | Exam Weight | Course Mapping | Core Engineering Competencies Tested |
|---|---|---|---|
| 1. AI Agents | 35% | Module 01Module 02Module 04 |
|
| 2. Prompt Engineering | 20% | Module 03Module 04 |
|
| 3. Data 360 Fundamentals | 20% | Module 03Capstone |
|
| 4. Testing, Deployment & Maintenance | 10% | Module 06Capstone |
|
| 5. Governance & Observability | 10% | Module 01Module 02Module 06 |
|
| 6. Multi-Agent Orchestration | 5% | Module 05Capstone |
|
Get Winter '27 Agent Script Patterns & AI-201 Study Notes
Join Salesforce Developers and Technical Architects receiving deep-dive tutorials on Agentforce Vibes IDE v4, Data 360 Hybrid RAG, MCP tool servers, and AI-201 certification blueprints. Instant single opt-in with one-click unsubscribe anytime.