Feb 13
Loading video player...
Custom agentic AI doesn’t have to be confusing. In this episode, we break down agentic AI architecture from the user interface all the way to tools, memory, LLM selection, and monitoring—so you can design a custom agent that actually works in production. Watch next: Agentic AI playlist for more episodes: https://www.youtube.com/playlist?list=PL82X3ahN92oxRu-lK-MLsmhfDFNhdjHEq Key takeaways ➡️ The end-to-end blueprint: how users interact, where agent logic runs, how tools connect, and how context is retrieved ➡️ The 6 core build decisions behind most custom agent stacks (UI, hosting, tools/MCP, memory/RAG, LLMs, monitoring) ➡️ A practical example: a Zendesk ticket triage agent that can act in Salesforce, use docs via RAG, store memory, and report results back Chapters 00:00 Why architecture matters 00:28 Why go custom vs platforms 01:57 Start with the interaction layer (UI / Slack / APIs) 04:32 Hosting the agent logic in cloud (AWS/GCP/Azure) 05:52 Tools + integrations (ETL, endpoints, services) 07:37 LLM choice: self-hosted vs API + multi-LLM routing 09:18 Monitoring and observability (usage, cost, health) 10:21 The six core decisions overview 11:16 Example architecture: Zendesk + Salesforce agent stack 15:56 Flexibility vs complexity + multiple front ends 18:58 Wrap Links & resources Companion architecture artifact: https://drive.google.com/file/d/1O3D0_vQ_nMmKzm5VdJOUkzwzfnhyKj6e/view?usp=sharing Gavin Franklin on LinkedIn https://www.linkedin.com/in/gavin-franklin-067247170/ Tim Harting on LinkedIn: https://www.linkedin.com/in/timothy-harting/ Cirrius Solutions https://cirriussolutions.com/contact/ Cirrius Blog: https://cirriussolutions.com/blog/ Apple Podcasts: https://podcasts.apple.com/ca/podcast/cirrius-talk/id1805880497 Spotify: https://open.spotify.com/show/1of2KEMIMDwzhv15MDlr0c Youtube: https://www.youtube.com/playlist?list=PL82X3ahN92owefWBRow-wFZeYBLJuKsz5 Greg Banks on LinkedIn: https://www.linkedin.com/in/greg-banks-9785197/ Jason Fowler Music: https://www.jasonfowlermusic.com/ Questions / follow-up: CirriusTalk@Cirriussolutions.com Music credit: Thanks to Jason Fowler for allowing us to use “Mystery Road” Guest info (from this episode) Tim Harting and Gavin Franklin discuss practical architecture decisions required to build custom AI agents across enterprise systems, including hosting, memory, RAG, MCP tools, and monitoring.