CGT Square Talk

CGT Square Talk - Agentic AI Explained: Models, Costs & Security Risks | Part 2

Anthony Roth & Armen Gasanyan

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 54:12

Part 2 of our AI series dives deeper into what’s actually powering agentic AI systems, the models behind the scenes.

In this episode, the team breaks down how different models are used across AI agents, why a model strategy is critical, and how cost, performance, and reliability all play into real-world implementations.

We also explore:

Why using a single model isn’t scalable
How fallback models prevent downtime
The real cost of APIs vs subscriptions
Why human-in-the-loop is still essential
And a major AI-driven security breach that highlights new risks companies face

Whether you're building with AI or just trying to understand how these systems actually work in production, this episode gives you a grounded, behind-the-scenes look.

Stay tuned for Part 3, where we’ll walk through a live demo and share practical tips & workflows.

00:00 – Intro & Recap of Part 1  
00:54 – What AI Models Actually Do (The “Brains”)  
02:01 – Why Model Strategy Matters  
03:30 – Cost Considerations & Token Limits  
05:02 – Primary vs Fallback Models Explained  
07:30 – Real Issues Switching Between Models  
09:44 – Trial & Error: Choosing the Right Model  
10:38 – When AI Gets It Wrong (Fake Sources Example)  
11:27 – Matching Models to Tasks (Research vs Coding)  
12:50 – Human-in-the-Loop: Why It’s Critical  
15:07 – AI in Development: Junior vs Senior Engineers  
16:47 – Where AI Fits in the SDLC  
18:48 – AI as a Tool, Not a Replacement  
20:32 – Strategy Before Implementation  
21:06 – Real Model Stack (What They Actually Use)  
23:49 – Subscription vs API Cost Breakdown  
26:01 – Real Example: $100 API Cost in One Day  
27:30 – Cheaper Model Alternatives (AWS, Open Models)  
28:54 – The Future of AI Cost Optimization  
32:54 – Enterprise vs Open Source Models  
34:33 – Compute Power & Infrastructure Realities  
35:59 – Why Design Matters Before Building AI  
37:39 – From Experimentation to Architecture  
40:20 – The “Skunkworks” Approach to AI  
41:40 – Major AI Security Risks (Real Breach Case)  
42:37 – How AI Can Exploit APIs  
44:25 – Why This Security Threat Matters  
46:55 – Is Software Security Falling Behind?  
49:59 – Why Traditional Pen Testing Isn’t Enough  
51:19 – AI vs Human Security Testing  
52:51 – What’s Coming in Part 3 (Demo + Tips)

CGT Square Talk
Hosted by: Anthony Roth & Armen Gasanyan 

New episodes every other Monday!