> Markdown version of [/videos/1902-behind-the-scenes-of-building-vs-code-harald-kirschner?t=274](https://www.wearedevelopers.com/videos/1902-behind-the-scenes-of-building-vs-code-harald-kirschner?t=274). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Behind the Scenes of Building VS Code - Harald Kirschner

Harald Kirschner claims AI has shifted the development bottleneck from writing code to reviewing it. Inside VS Code, this multi-agent acceleration forces engineers to act as software architects.

- **Speakers:** Harald Kirschner
- **Event:** Coffee With Developers
- **Published:** June 10, 2026
- **Duration:** 32:38
- **URL:** https://www.wearedevelopers.com/videos/1902-behind-the-scenes-of-building-vs-code-harald-kirschner

## Summary

Modern developer environments are undergoing a massive shift as AI tools transition from simple assistants to autonomous workflow engines. The VS Code architecture, inherently web-based with HTML, CSS, and JavaScript, enables deeply integrated tooling where AI agents can natively reproduce UI issues and automate testing via Playwright. As development accelerates, bottlenecks have shifted from code generation to continuous integration and code review. To sustain multi-agentic parallel workflows, engineering teams must highly optimize underlying infrastructure, illustrated by the structural decision to rewrite the TypeScript compiler in Go to unlock a 10x reduction in build times rather than relying on Rust-backed approaches that do not address specific threading bottlenecks.

This accelerated output fundamentally changes how open-source projects and internal teams operate. Projects are increasingly flooded with low-friction, AI-generated pull requests and overly lengthy feature requests. To combat this noise, developer workflows must adopt a product-management mindset, emphasizing crisp problem definition and automated issue triage over mere code execution. Successful developers now function more like software architects, utilizing AI for rapid, disposable prototyping and feasibility designs rather than production-ready codebase generation. By focusing on high-level one-pagers and clear intent, engineers can prompt agents to iterate collaboratively and challenge assumptions, resulting in more robust system design.

To manage complex tasks without exhausting token limits, modern frameworks employ a hierarchical agent architecture that cleanly separates concerns. A main orchestration agent for a planning or test-driven development (TDD) loop delegates highly defined tasks to sub-agents. These sub-agents operate in isolated contexts, utilizing faster, cheaper models like Haiku or GPT-4 Mini to crawl file trees or validate tests. By returning only summarized results rather than raw data, the framework achieves strict context isolation and parallel processing. This multi-agent scaling drastically reduces token burn while shielding top-level models from context pollution during intricate software compilation.

**Keywords:** vs code architecture, multi-agentic parallel workflows, UI issue testing agents, typescript compiler optimization, CI/CD infrastructure bottlenecks, hierarchical sub-agent processing, token context isolation, context window management, TDD loop automation, open source PR triage, automated issue template engineering, agentic architectural prototyping, LLM token burn reduction, multi-agent orchestration, developer bottleneck shifting

## Chapters

1. **Leveraging web-based architecture for custom text editor features** (00:01) — Running a browser environment inside a text editor enables deeper customization and visual debugging through familiar web dev tools.
1. **Transitioning from direct code editing to multi-agent orchestration** (02:44) — Modern workflows shift focus from writing code to reviewing output and orchestrating parallel tasks across specialized development agents.
1. **Rewriting core compilation tools to resolve concurrent execution bottlenecks** (04:34) — Migrating the typescript compiler to go dramatically reduces build times and prevents parallel testing agents from stalling the continuous integration pipeline.
1. **Offloading computational pressure from local workspace language servers** (07:27) — Moving heavy repository operations and agent workloads into isolated processes stabilizes local integrated development environments.
1. **Enforcing automated code reviews to manage accelerated delivery** (08:25) — Requiring artificial intelligence review clearance before human evaluation prevents fast-moving generative pull requests from overwhelming primary maintainers.
1. **Filtering overwhelming community generative submissions in open source** (10:23) — Reversing the submission pipeline to prioritize validated issue discussions over automated multi-page pull requests preserves project maintainability.
1. **Structuring complex problem reports using automated conversation templates** (13:40) — Deploying conversational assistants to clarify ambiguous technical issues prevents lengthy unstructured essays from stalling developer triage queues.
1. **Abstracting routine product tasks into dynamic instructional prompts** (16:04) — Configuring custom natural language skill endpoints automates repetitive documentation chores and operational workflows within standard chat interfaces.
1. **Securing vulnerable generation environments against malicious marketplace extensions** (19:14) — Implementing rigorous prompt scanning and trusted publisher curation protects developers from installing corrupted workflow plugins that leak local data.
1. **Steering generative output using high-level architectural vision documents** (21:54) — Utilizing concise single-page specifications allows software architects to establish strong conceptual constraints that keep generation models aligned with broader goals.
1. **Accelerating early conceptual development via disposable exploratory generation** (26:03) — Treating initial code concepts strictly as disposable prototypes prevents teams from getting permanently locked into suboptimal incremental architecture fixes.
1. **Isolating logic sequences using parallel automated generation routines** (29:18) — Spawning smaller specialized routines allows base systems to quickly evaluate code constraints in isolated contexts without polluting global memory.

## Related Moments

- [Rethinking team structures around AI agent capabilities](https://www.wearedevelopers.com/videos/1539-agentic-devops-how-ai-powered-automation-transforms-software-delivery-on-github-and-azure) (from "Agentic DevOps: How AI-Powered Automation Transforms Software Delivery on GitHub and Azure")
- [Shifting developer workloads and realistic AI productivity gains](https://www.wearedevelopers.com/videos/1830-wearedevelopers-live-speculaitions) (from "WeAreDevelopers LIVE - SpeculAItions")
- [Building a native AI agent experience within the IDE](https://www.wearedevelopers.com/videos/1401-the-alpha-developer-of-tomorrow-building-the-future-of-the-software-development-lifecycle) (from "The Alpha‑Developer of Tomorrow: Building the Future of the Software Development Lifecycle")
- [Replacing traditional code reviews with interactive agents](https://www.wearedevelopers.com/videos/100353-meta-harnesses-what-the-js-framework-wars-teach-us-about-what-is-next-for-ai-agents) (from "Meta Harnesses: What the JS Framework Wars Teach Us About What Is Next for AI Agents")
- [Transitioning into the integrated era of agentic IDEs](https://www.wearedevelopers.com/videos/2097-ai-code-then-vs-now-from-complex-rubbish-to-co-piloting-in-12-months) (from "AI Code then vs now: From Complex rubbish to co-piloting in 12 months")
- [Embracing code generation and modern editor evolution](https://www.wearedevelopers.com/videos/1403-five-things-in-tech-that-matter-and-we-have-to-make-work) (from "Five things in tech that matter and we have to make work")

## Related Articles

- [What is Agentic Programming and Why Should Developers Care?](https://www.wearedevelopers.com/magazine/625-what-is-agentic-programming-and-why-should-developers-care)
- [Never delegate the understanding](https://www.wearedevelopers.com/magazine/749-never-delegate-the-understanding)
- [How we Build The Software of Tomorrow](https://www.wearedevelopers.com/magazine/120-how-we-build-the-software-of-tomorrow)
- [Transforming Software Development: The Role of AI and Developer Tools](https://www.wearedevelopers.com/magazine/527-transforming-software-development-the-role-of-ai-and-developer-tools)

## Related Jobs

- [Principal Product Manager, Agent Platform](https://www.wearedevelopers.com/jobs/ext/277541-principal-product-manager-agent-platform) at **GitHub**
- [Staff Software Engineer, Copilot Experiences](https://www.wearedevelopers.com/jobs/ext/164361-staff-software-engineer-copilot-experiences) at **GitHub**
- [Senior AI Agent Software Engineer (Go, Python) (m/f/x)](https://www.wearedevelopers.com/jobs/48277-senior-ai-agent-software-engineer-go-python-m-f-x) at **Dynatrace**
- [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat**
- [Senior Software Engineer](https://www.wearedevelopers.com/jobs/ext/15942-senior-software-engineer) at **GitHub**
- [Principal Software Engineer, Enterprise AI Platform](https://www.wearedevelopers.com/jobs/ext/1467292-principal-software-engineer-enterprise-ai-platform) at **GitHub**