AI Realist visits with Haverin: Reality Bites! (New Podcast episodes)
The AI Expert Problem Has a REALIST Antidote. Her Name is Maria Sukhareva.
Maria Sukhareva was working in natural language processing before AI was even a glimmer in some peoples eye.
She built the first NLP teams at BMW and Siemens, has a computational linguistics background predating the transformer era, and writes the AI Realist Substack, where she cuts through the hype with the kind of technical grounding that only comes from actually ‘been there, done that’.
We chatted for almost 2 hours on a Saturday. I carved that up into three episodes, and we covered a LOT of ground.
If you follow Maria, you know what you're in for, and it’s great.
If you don't, start here:
AI Realist Part 1 – Origins
Before ChatGPT Made Everyone an AI Expert: Maria’s Story
Maria traces the direct line from Bible translation corpora to modern LLMs, explains why her linguistics background gives her a fundamentally different read on what these models are actually doing, and tells the story of hiding the BMW NLP team inside the blockchain department just to get access to GPUs.
How machine translation became the foundation for large language models
Why the “overnight AI expert” phenomenon bothers people who were doing this work in 2018
What it actually means when an LLM predicts its next word — and why that matters for how you use them
YOUTUBE
🎙️ APPLE PODCASTS
🎵 SPOTIFY
AI Realist Part 2 – Reality
2026: GPT-5, Claude, AI Agents and ROI: Maria Sukhareva on What’s Actually Real
Maria and I disagree on almost nothing, which made this the most useful episode. GPT-5, the ROI gap, vibe coding, why the internet is turning purple, and the only honest test for whether an AI use case actually matters.
Why Maria called GPT-5 an incremental improvement — and what happened to the “visionary” experts afterwards
The one question that separates real AI impact from nice-to-have: what happens if it goes down tomorrow?
Coding, model collapse, and why we’re already seeing the snake eat its tail
YOUTUBE
🎙️ APPLE PODCASTS
🎵 SPOTIFY
AI Realist Part 3 -Horizons
Digital Twins, Open Weights and Why China Could Win AI
The episode where we stopped being polite about what’s actually being delivered. Digital twins, agentic AI, China’s open-weight advantage, and the single most practical piece of advice either of us has offered in this space.
Why most “digital twins” aren’t — and my four-level maturity model for what a real one would require
Why most “autonomous agents” are a system prompt connected to a SharePoint folder
Maria’s three-day rule: build one before you buy one, sell one, or put one on a slide
🎙️ APPLE PODCASTS
🎵 SPOTIFY
All three episodes are live now.
Maria’s AI Realist Substack. If you work anywhere near healthcare technology, enterprise software, or AI strategy, and you’re not already reading her, fix that. Subscribe NOW!
Hey — I came across your writing and really liked how you think.
I’m exploring something similar from a different angle — writing about human behavior through a system design lens (like debugging internal patterns).
Just started publishing on Substack. If you ever get a moment to read, I’d genuinely value your perspective.
Also happy to support your work — feels like there’s an interesting overlap here.