Architecture
See how the editor, agents, knowledge base, and repositories exchange context without collapsing into one undifferentiated folder.
Building an AI Engineering Workspace
An AI engineering workspace should be more than an editor with a chatbot attached. It should give humans and agents a shared place to work, preserve the reasoning behind changes, and make the next session easier than the last.
This section documents the workspace I am building around VS Code, agent integrations, Git repositories, and a Markdown knowledge base that remains useful in Obsidian. It is a working system, not a claim that one folder structure will suit everyone.
The central idea
Section titled “The central idea”Each tool has a clear responsibility:
- VS Code is the primary workbench for editing, review, terminals, and repository context.
- AI agents investigate, implement, verify, and maintain work within explicit boundaries.
- Markdown carries durable project knowledge without locking it inside a particular application.
- Obsidian provides a human-friendly view across notes, decisions, projects, and relationships.
- Git and GitHub preserve code history and provide deliberate publication and review boundaries.
The value comes from the handoffs between them—not from any tool in isolation.
Working workflows
Follow practical patterns for starting projects, research, agent handoffs, and recording decisions.
Agent onboarding
Give a new agent enough context to act without paying to reload the entire workspace into every session.
Landmark documents
Preserve expensive debugging and architecture context close to the implementation that depends on it.
Frontmatter and trust
Give humans and agents a compact way to judge what a note is, how current it is, and what evidence supports it.
What this section will become
Section titled “What this section will become”The documentation will evolve alongside the workspace. As useful conventions survive real projects, they will be recorded here with their tradeoffs. Failed approaches belong here too when they explain why the current system exists.
The goal is not maximum automation. It is a workspace that remains understandable, recoverable, and productive when the human or agent working in it changes.