Show notes
RedMonk's James Governor sits down with Mina Ilieva, AI engineer at TurinTech AI, to talk about a problem everyone in 2026 recognizes: we're shipping a lot of code slop. Ilieva explains how TurinTech's platform, Artemis, fights back. It began life running a genetic algorithm that scores candidate code against a fitness function, and it now has a newer tool, Discovery, built on an empirical loop where agents form hypotheses, turn them into experiments, and verify every change before a human approves it. The two get into what clients actually optimize for — throughput, latency, memory, runtime — and why none of it is free. Ilieva walks through real wins with Intel's vLLM work, QuantLib, BLAKE3, and a quantized Nemotron model. They also take apart "token maxxing," the habit of burning tokens to look busy, and make the case that verification skill, not raw output, is what keeps engineers employable.This RedMonk conversation is sponsored by TurinTech AI.Show notes: https://redmonk.com/videos/mina-ilieva/Chapters00:00 Introduction to AI and TurinTech02:52 Code Optimization and Its Importance05:35 Use Cases and Client Engagements08:55 Token Efficiency and Cost Management11:46 The Role of AI in Software Development14:47 Trade-offs in Optimization17:37 Future of AI and Model Deployment