Episode 17 · Season 1
Agents, MCPs, and Skills: The 3 Building Blocks of AI Agents Explained
What if your AI agents were architected as intentionally as your Salesforce org? In this episode of Cirrius Talk, Greg Banks sits down with Tim and Gavin to break down a practical mental model for AI agent architecture, agents as decision-makers, MCP servers as shared capabilities, and skills as reusable workflows. They unpack when a use case calls for a new skill, a new MCP, or a brand-new agent (or even a multi-agent pattern), plus the anti-patterns to avoid as you scale.
What if your AI agents were architected as intentionally as your Salesforce org? In this episode of Cirrius Talk, Greg Banks sits down with Tim and Gavin to break down a practical mental model for AI agent architecture—agents as decision-makers, MCP servers as shared capabilities, and skills as reusable workflows. They unpack when a use case calls for a new skill, a new MCP, or a brand-new agent (or even a multi-agent pattern), plus the anti-patterns to avoid as you scale. By the end, leaders walk away with questions they can use to design a secure, future-ready AI workforce across their Salesforce ecosystem.
Takeaways
- Agents are the “brains,” not the plumbing—let them decide what to do when, not handle low-level integration.
- MCP servers are your superpowers: secure, reusable connections into Salesforce, Jira, ERPs, and document stores that every agent can tap.
- Skills are where your workflows live—think “mini playbooks” that orchestrate steps and call MCPs to actually get work done.
- Use the five-question test (audience, decision space, context, security, triggers) to know when a new agent is justified.
- Multi-agent patterns shine when a “hub” agent needs to reason across multiple domain-specific agents—sales, service, finance, delivery, and more.
- Build from the bottom up (capabilities and MCPs), but design from the top down (business decisions and outcomes).
- Cirrius Solutions brings AI-shoring to Salesforce: 100% US-based teams, 50% military veterans and spouses, all AI & Data Cloud certified.
Recommended Links
Send questions to cirriustalk@cirriussolutions.com
We cover
- Understanding AI Agent Architecture Terminology
- Exploring MCPs and Skills
- Deciding Between Separate Agents and Multi-Agent Patterns
- Holistic Approach to AI Agent Architecture
Chapters
- 0:00Welcome Back to Tim and Gavin
- 1:09 Understanding AI Agent Architecture Terminology
- 8:52Exploring MCPs and Skills
- 12:16Deciding Between Separate Agents and Multi-Agent Patterns
- 14:46Holistic Approach to AI Agent Architecture
Resources and downloads
The artifacts Greg’s team built for this episode, free to download.
Agents, MCPs, and Skills: The 3 Building Blocks of AI Agents Explained
A one-page visual mental model for deciding whether a given capability belongs in an MCP, a Skill, or an Agent, and when to split one agent into several. Includes the decision questions for each layer, guidance on separate agents versus multi-agent designs, and three anti-patterns that make agent architectures hard to share and version.
Related episodes
Agentic AIAgentforce vs Custom Agents: The Real Cost Breakdown (With Numbers)
Agentic AIFrom Clicks to Code: How AI Is Redefining Salesforce Development
Agentic AIDreamforce Debrief: AI Agents, Slack Smarts, and the Changing Consulting Game
The Military MinuteTom House – Director of Salesforce Military: Serving Those Who ServedTom House, Director of Salesforce Military
The Military MinuteJason Pomeroy – VeteranJason Pomeroy, VeteranTalk to us about AI agents
Keep going: See how we build AI agents for real businesses, Talk to us about a custom AI agent, More from the Cirrius blog