Uploaded November 2026 | Updated November 2026, 3 hours ago
Stop typing or copy-pasting 100-line prompts into your AI coding agent every session!
In this episode, Herbert Wolverson covers how to persist project rules, stack definitions, and architecture constraints using AGENTS.md (and CLAUDE.md) —and how to avoid the dangerous trap of context window bloat and token degradation.
Herbert will look at how to structure generic rules, extend existing APIs in under 2 minutes, analyze where your LLM context window actually goes, and find the right trade-off between guidance and performance.
🔥 WHAT YOU'LL LEARN:
• Moving beyond 100-line prompts: How to persist rules across sessions and team members
• Setting up AGENTS.md & CLAUDE.md: Defining tech stacks, architectural rules, and linting guidelines
• Real-World Demo: Adding a new REST endpoint in 1 min 42 secs using OpenCode
• The Danger of Context Window Bloat: Why adding entire books or style guides causes context compacting and AI quality degradation
• Where Context Goes: How system prompts, tool definitions (MCP, Playwright), nested agents files, and chat history consume your token budget
• Engineering Trade-offs: Knowing what to put in AGENTS.md vs. relying on pre-trained model knowledge
💡 NEXT UP IN THE SERIES: Herbert will explore coding harness skills files and how to smartly manage context without overloading your session!
---
Explore more from Ardan Labs
Online Courses: ardanlabs.com/education
Live Training Events: ardanlabs.com/live-training-events
Technical Blog: ardanlabs.com/blog
Github: github.com/ardanlabs
---
Connect with Ardan Labs
Website: ardanlabs.com
X: https://x.com/ardanlabs
LinkedIn: linkedin.com/company/ardanlabs
#aicoding #AgentsMD #claudecode #opencode #softwareengineering #rustlang #llm #developertools
Stop typing or copy-pasting 100-line prompts into your AI coding agent every session!
In this episode, Herbert Wolverson covers how to persist project rules, stack definitions, and architecture constraints using AGENTS.md (and CLAUDE.md) —and how to avoid the dangerous trap of context window bloat and token degradation.
Herbert will look at how to structure generic rules, extend existing APIs in under 2 minutes, analyze where your LLM context window actually goes, and find the right trade-off between guidance and performance.
🔥 WHAT YOU'LL LEARN:
• Moving beyond 100-line prompts: How to persist rules across sessions and team members
• Setting up AGENTS.md & CLAUDE.md: Defining tech stacks, architectural rules, and linting guidelines
• Real-World Demo: Adding a new REST endpoint in 1 min 42 secs using OpenCode
• The Danger of Context Window Bloat: Why adding entire books or style guides causes context compacting and AI quality degradation
• Where Context Goes: How system prompts, tool definitions (MCP, Playwright), nested agents files, and chat history consume your token budget
• Engineering Trade-offs: Knowing what to put in AGENTS.md vs. relying on pre-trained model knowledge
💡 NEXT UP IN THE SERIES: Herbert will explore coding harness skills files and how to smartly manage context without overloading your session!
---
Explore more from Ardan Labs
Online Courses: ardanlabs.com/education
Live Training Events: ardanlabs.com/live-training-events
Technical Blog: ardanlabs.com/blog
Github: github.com/ardanlabs
---
Connect with Ardan Labs
Website: ardanlabs.com
X: https://x.com/ardanlabs
LinkedIn: linkedin.com/company/ardanlabs
#aicoding #AgentsMD #claudecode #opencode #softwareengineering #rustlang #llm #developertools










