VS Code Insiders Podcast Inside The Agent Loop with Pierce Boggan
9 snips
Apr 20, 2026 
Pierce Boggan, a Visual Studio Code engineer who builds the agent loop and agentic code workflows. He explains how agent loops, sub-agents, tools, and harnesses interact. You’ll hear why different models are picked for specific tasks. The conversation covers prompt construction, tool schemas, context limits, and how to customize or disable sub-agent behaviors.
AI Snips
Chapters
Transcript
Episode notes
Agent Loop Is An Iterative Tool-Driven While Loop
- The agent loop is a repeated API-driven while loop combining system prompts, explicit/implicit context, tools, and the user prompt to iteratively solve tasks.
- Pierce explains the loop: model picks tools (search, read, edit), executes them, appends results, and iterates until it returns a stop message to the user.
Context And Tools Are Appended Per Request
- VS Code supplies both explicit context (file paths, terminal state) and implicit context (open editor, environment) appended to every model request to improve relevance.
- Tools are provided with schemas; the model fills parameters and VS Code executes actions like search, read, or edit.
Limit Prompt Bloat By Using Tools And Skills
- Avoid dumping everything into prompts; use tools and skills strategically to prevent context overflow and decision fatigue for the model.
- Pierce notes models struggle with too many choices and the team builds custom models to refine tool lists per session.