Behind the Scenes of Building VS Code - Harald Kirschner

What if your job wasn't to write code, but to orchestrate AI agents? The VS Code team shares how they are already building software this way.

#1about 2 minutes

Understanding the web-based architecture of VS Code

VS Code is built on web technologies like Electron, which enables powerful meta-debugging capabilities and an accessible extension ecosystem.

#2about 3 minutes

How AI agents are changing developer workflows

Development is shifting from manual code editing to reviewing AI-generated code and orchestrating agents, creating new bottlenecks in CI/CD pipelines.

#3about 2 minutes

Boosting build performance with a TypeScript to Go rewrite

The TypeScript compiler was rewritten in Go to achieve a 10x performance improvement, addressing critical build time bottlenecks for the VS Code team.

#4about 4 minutes

Adapting the SDLC for faster, agent-assisted releases

To support faster agent-driven development, the team implemented mandatory AI code reviews and automated triage to maintain quality throughout the entire release cycle.

#5about 4 minutes

Managing the influx of AI-generated open source contributions

Open source projects are combatting low-quality AI-generated pull requests by requiring contributors to first file well-defined issues and plans for discussion.

#6about 6 minutes

Building and securing custom agentic workflows

While custom agentic workflows can be created with natural language, users must be cautious of security risks from malicious extensions and skills in marketplaces.

#7about 5 minutes

The evolving role of developers and architects in the AI era

Individuals with both a technical background and product sense are most successful with AI, as they can guide agents with high-level vision rather than overly detailed specifications.

#8about 3 minutes

Leveraging AI for rapid prototyping and idea exploration

Use AI to create rapid prototypes for team feedback and idea validation, which is more effective than jumping directly to a final implementation.

#9about 3 minutes

How sub-agents enable context isolation and parallelism

Sub-agents are spawned by a main agent to perform isolated tasks, allowing for parallel processing and context window optimization by using cheaper, faster models.