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

#1115 - FFAF: Matt Walsh and I Agree on When Our Culture Died

The Counsel of Trent

When did our culture die?

The AI-powered Podcast Player