VS Code Insiders Podcast Models, Evals, and Raptor Mini with Julia Kasper
30 snips
Dec 8, 2025 Julia Kasper, an engineer focused on AI integrations for VS Code, discusses the transformative power of AI in developer workflows. She explains how model evaluation and fine-tuning enhance coding experiences. Julia highlights the significance of Raptor Mini for quick tasks and the nuances of choosing and adapting models. The conversation also dives into model changes over time and the importance of feedback in refining AI tools. Listeners gain insights into maximizing efficiency with cutting-edge technologies!
AI Snips
Chapters
Transcript
Episode notes
AI Changes How Developers Work
- Julia Kasper says AI is reshaping developer workflows by speeding up development and blending low-code feel with visible code control.
- She finds AI accelerates tasks while preserving visibility into what's happening under the hood.
Continuously Reevaluate Model Choice
- Re-evaluate which model you use for each task instead of sticking to one by default.
- Try different models over time because providers and client prompts are continuously changing.
Client Prompts Drive Perceived Model Changes
- VS Code modifies client-side prompts before calling model endpoints and updates them from feedback and offline evaluation.
- These prompt changes can make a model feel different from week to week even without provider changes.