Software engineers often think adding AI is just a simple API call, but moving from a Proof of Concept to a stable production system requires a completely different mindset.
Maria Vechtomova breaks down the harsh reality of MLOps, why rigorous evaluation is non-negotiable, and why autonomous agents are riskier than you think.
In this episode, we cover:
- The essential MLOps principles every software engineer must learn
- How to bridge the gap between a demo and a production-grade solution
- Strategies for evaluating agents and detecting model drift
- The security risks of customer service agents and prompt injection
- Practical tips for using AI tools to boost your own productivity
Connect with Maria:
https://www.linkedin.com/in/maria-vechtomova
Timestamps:
00:00:00 - Intro
00:01:25 - Why the AI Hype Was Actually Good for Monitoring
00:03:07 - Real-World AI Use Cases That Deliver Actual Value
00:05:16 - MLOps Basics Every Software Engineer Needs to Know
00:08:08 - The Hidden Complexity of Deploying Agents to Production
00:12:02 - Minimum Requirements for Moving from PoC to Production
00:15:41 - Step-by-Step Guide to Evaluating AI Features Before Launch
00:18:08 - How to Handle Data Labeling and Drift Detection
00:21:55 - Why You Likely Need Custom Tools for Monitoring
00:24:56 - Why Engineers Build AI Features They Don't Need
00:26:01 - How Software Engineers Can Learn Data Science Principles
00:31:36 - The Dangerous Security Risks of Autonomous Customer Service Agents
00:34:44 - Why Human-in-the-Loop is Essential for Avoiding Reputational Damage
00:36:18 - Boosting Developer Productivity with Opinionated AI Prompts
00:39:20 - Using Voice Notes and AI to Organize Your Life
#MLOps #SoftwareEngineering #ArtificialIntelligence
