Same platform. Same Atlas Reasoning Engine. Same Trust Layer underneath. So why do employee agents and customer-facing agents need completely different architecture? In this episode, Misbah and Frankie draw the real line not "internal vs. external," but "how fast and how visibly does the consequence show up" and tell the true story of an internal sales assistant that confidently recommended an unapproved discount to a rep, who then promised it to a real customer.

One agent trying to be an expert in returns, orders, and loyalty ends up mediocre at all three. In this episode, Misbah and Frankie (playing two roles) explore Multi-Agent Orchestration — how specialized Agentforce agents collaborate through open standards like Agent2Agent and MCP and tell the true story of two perfectly-designed specialist agents that got stuck in a perfectly polite, perfectly infinite loop, each waiting on the other for information neither could provide first.

Grounding sounds like magic until you ask the real question: grounded against what? In this episode, Misbah and Frankie dig into how Data 360 and Customer 360 give Agentforce a single, unified truth to reason against, including the shift toward zero-copy, federated grounding and tell the true story of a proactive personalization agent that confidently, correctly grounded itself against quietly broken data, and sent goth streetwear recommendations to a loyal petite-sizing customer.

Grounding. Data masking. Toxicity detection. Zero data retention. Audit logging. The Trust Layer isn't one feature, it's five specialists standing guard at five different checkpoints between your model and your customer. In this episode, Misbah and Frankie open the velvet rope on what the Trust Layer actually does, why it doesn't replace your existing Salesforce security model, and the true story of the night a well-meaning developer quietly talked the safety feature out of doing its job — and almost shipped the wrong benefits answer to every employee at Frankencare.

Two Subagents is easy. Thirty-one Subagents with four of them named some version of "General" is a disaster waiting to confuse both your customers and your on-call engineer.

In this episode, Misbah and Frankie lay out the naming conventions, reuse patterns, and risk classifications that keep Subagents, Actions, and Instructions clean as an Agentforce org scales and then walk through the true story of a real 31-Subagent cleanup, refactored live down to 11.

Most failed Agentforce builds don't fail because a Subagent was misconfigured, they fail because nobody could answer "what is this agent actually responsible for" before anyone opened Agent Builder. In this episode, Misbah and Frankie borrow the CTA mindset, capability first, boundaries second, implementation third and introduce the "Agent Capability Contract," a five-part design document every architect should write before touching a single Subagent. Plus: the story of the Frankencare committee that couldn't answer three simple questions, and what happened when they finally could.

Everyone thinks Agentforce is just a chatbot with a Salesforce badge. It isn't and the gap between those two ideas is exactly where production incidents are born. In the season opener of Season 3 - The Agentforce Architects, Misbah and Frankie break down the six-step Atlas Reasoning Engine loop and the three levers every architect controls: Subagents, Actions, and Instructions. Then things get dramatic: the true(ish) story of the night Frankencare's first customer service agent approved 47 refunds before sunrise followed by the replay of the same day, done right.

🎧 New episodes every week. 🧟 Subscribe so you don't miss the next chaos-agent story.

Frankie realizes, mid-episode, that he's somehow been carrying master keys to every room in the castle , rooms he has no business being in. A deep dive into least privilege, profiles, and the permission sprawl that quietly builds up over years.