AI agents don’t fail because they’re “not smart enough”—they fail because they’re missing the right context. In this episode of Cirrius Talk, Greg Banks sits down with Tim and Gavin to break down context engineering: the practical evolution beyond prompt engineering that’s shaping modern agent performance. You’ll learn the five core context sources agents draw from (prompts, system instructions, memory, RAG, and external tools) and why smaller, curated context windows often produce the most accurate results. If you’re building agents in the Salesforce ecosystem—or anywhere—this is your blueprint for making them faster, safer, and more reliable. 

Highlights / Takeaways

• 🧠 Context > prompts: prompt engineering evolved—now it’s about managing what the agent knows
• 🧱 5 context types to design for: prompts, system instructions, memory, RAG, and external tools
• 🎯 Less is more: bloated context increases hallucinations and misfires—curate ruthlessly
• 🗂️ RAG is your process-at-scale: SOPs, checklists, and guidelines become agent-ready knowledge
• 🛠️ Tools unlock real work: CRMs, ERPs, weather APIs, and databases turn chat into outcomes

Recommended Links

Episode Artifact

Gavin Franklin on LinkedIn

Tim Harting on LinkedIn

Greg Banks on LinkedIn

Cirrius Solutions

Cirrius Blog

Mental Model Diagram

Jason Fowler Music

Send questions to [email protected]