Maria Sukhareva | AI Realist: Part 2 – Reality (AI 2026: GPT-5, Claude, AI Agents and ROI: Maria Sukhareva on What’s Actually Real)

The Reality of AI: GPT-5 "Flops," Vibe Coding, and the ROI Crisis

Is the AI hype cycle finally crashing? In this episode of the Haverin About podcast, Stuart Miller sits down with AI expert and "Realist" author Maria Sukhareva to peel back the curtain on what’s actually happening in the world of Large Language Models (LLMs).

From the perceived "flop" of GPT-5 to the rise of "Vibe Coding," we explore why many AI agents over-promise and under-deliver. If you are a business leader wondering where the Return on Investment (ROI) is for AI, or a developer trying to navigate the "snake eating its own tail" in generated code, this deep dive is for you.

In This Episode, We Discuss:

• The GPT-5 Turning Point: Why Maria believes GPT-5 was an incremental improvement at best and what it means for the "visionary" experts.
• The ROI Gap: Why 25,000 "AI agents" at McKinsey might not be as impactful as they sound.
• Vibe Coding & Productivity: How AI-assisted coding is the only "transformative" use case right now—and the risk of "boring" code and "purple websites".
• Claude vs. ChatGPT: Why Stuart switched to Claude for long-form writing while keeping ChatGPT for research.
• Digital Twins in Healthcare: Moving from engineering models to the "House" diagnostician concept for patient care.
• The Small Model Debate: Why Maria is a "small model skeptic" but sees hope in Chinese innovations like Minimax.

Timestamps:

0:00 – Introduction: The AI Hype Cycle vs. Reality

0:35 – Did GPT-5 Flop? The Shift to AI Skepticism

2:34 – Where is the ROI? Chatbots vs. Useful Agents

4:11 – The "Tulip Investment" Market of the 18th Century

6:19 – What is Vibe Coding?

8:53 – The Snake Eating Its Tail: Model Collapse in Code

11:27 – Why is the Internet Turning Purple?

13:12 – The 5-Model Website Building Experiment

16:18 – Why ChatGPT Leads to "Writing Hell"

18:12 – The Trouble with Prompt Engineering & Attention

20:15 – AI in Healthcare: Interrogating Medical Records

28:20 – The Future of Digital Twins & Master Diagnosticians

34:21 – Large vs. Small Language Models: The Generalization Problem

39:01 – Local AI: Running Minimax on your Desk

Connect with Us:

• Read Stuart Miller on Substack (Haverin): https://haverin.substack.com
• Subscribe to Maria Sukhareva (The AI Realist): https://airealist.substack.com
• Follow Maria Sukhareva on LinkedIn: https://www.linkedin.com/in/msukhareva/
• Follow Stuart Miller on LinkedIn: https://www.linkedin.com/in/stuart-miller-haverin/

Keywords: GPT-5, AI Agents, Vibe Coding, AI ROI, LLM Comparison, Claude vs ChatGPT, Digital Twins Healthcare, Maria Sukhareva, Stuart Miller, AI Assisted Coding, Mixture of Experts, Small Language Models.