EP 03 — The Agentforce Architecture Overview

Every Agentforce agent — no matter how complex — is built from the same five parts. In Episode 3 of the Agentforce Complete Learning Series, we break down the full architecture: Agent Definition, Topics, Actions, Instructions, and Channels, plus the Atlas Reasoning Engine that ties it all together. We start with a real scenario — a customer who's lost their credit card — and walk it through the entire reasoning loop (Receive → Classify → Reason → Act → Observe → Respond) so you can see the architecture in action before we name any parts. Then we break down each of the five architectural areas one at a time, meet Atlas (the reasoning engine behind every decision), cover where channels fit in, and clear up the most common misconceptions about how Agentforce actually works under the hood. 📌 CHAPTERS 0:00 The Agentforce Architecture Overview (Intro) 0:48 Where We Are in the Series 1:20 What You Will Know After This Episode 1:51 Before We Name the Parts — Watch It Work 3:00 Think of It Like Hiring a New Employee 4:04 Five Architectural Areas of an Agentforce Implementation 5:42 Agent Definition — The Identity Layer 6:51 Topics / Subagents — Specialised Jobs and Classification 8:17 Actions — What the Agent Actually Does 9:56 Meet Atlas — The Brain Behind Every Decision 11:23 The Agentforce Reasoning Loop (Receive → Classify → Reason → Act → Observe → Respond) 12:46 Channels — Where Agents Engage 13:55 What Agentforce Architecture Is Not (Common Misconceptions) 14:50 Now You Know What Ran (Scenario Recap) 16:55 You Now Have the Complete Blueprint (Summary & What's Next) 🔔 Catch up on the series: Episode 1 (What is Agentforce?) and Episode 2 (How AI Actually Thinks Inside Salesforce) before this one if you haven't already. Subscribe so you don't miss Episode 4 — Standard vs. Custom Agents. 💬 Which architectural piece do you want a deeper dive on — Topics, Actions, or the Atlas reasoning loop? Tell us in the comments. AstraAltus — Learn. Build. Lead.