← back
Vibe Coding at Scale: Customizing AI Assistants for Enterprise Environments - Harald Kirshner,
Takeaway
Enterprise vibe coding works when you graduate from YOLO mode to templates plus MCP-based tools that constrain agents to your stack and conventions.
Summary
- Three stages of vibe coding: YOLO vibes (prototyping/personal), structured vibes (template + guard rails), spec-driven vibes (scaled enterprise codebases).
- VS Code Insiders demoed: agent mode, auto-approve, built-in local-model voice dictation, new-workspace tool, browser preview with element selection, auto-accept-after-delay, undo/revert.
- Model comparison anecdote: Claude Sonnet 4 best at polished UI, GPT-4.1 wireframey, Gemini 2.5 Pro varied, others tend toward 'blockchain landing page' aesthetic.
- Templates with consistent stacks (React+Vite+Material UI etc.) plus repository instructions give non-technical contributors guard rails.
- Custom tools via MCP elevate reliability and consistency by injecting context instead of relying on YOLO inference.
vibe-codingvs-codegithub-copilot
Original description
"Vibe coding" often falters in complex enterprise environments. Drawing from real implementations, this talk demonstrates systematic approaches to customizing AI assistants for challenging codebases. We'll explore specialized techniques for navigating complex architectures, evidence-based strategies for undocumented legacy systems, methodologies for maintaining context across polyglot environments, and frameworks for standardizing AI usage while preserving developer autonomy. Through case studies from finance and healthcare, we'll present a comprehensive evaluation framework that bridges the gap between AI's theoretical capabilities and practical enterprise implementation, enabling true flow-state collaboration even within the most complex development ecosystems. About Harold Kirshner I'm Harald Kirschner, a Principal Product Manager at Microsoft working on Visual Studio Code and GitHub Copilot, supporting over 40 million active developers code faster and more efficiently across virtually any programming language. Before Microsoft, I led Developer Experience at Mozilla, where I led Firefox DevTools and helped deliver Firefox Quantum, which doubled browser performance. My background in software engineering, including early work on MooTools, gives me hands-on insight into the challenges developers face daily. When I'm not working, I enjoy hiking California's coastal trails and experimenting with generative art. As a speaker at the AI Engineer Summit, I'm excited to share insights from our work on AI coding tools and Model Context Protocol to help developers achieve flow state even in complex environments. Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter