Research
Context Minimization Principle
The core theory in the Context Cost research program.
Explore CMPContext Cost
Insights on software engineering in the AI era.
Research and writing on software design, context cost, and AI-assisted engineering.
By Leric ZhangExplore
Writing
Essays
Short arguments on software architecture, AI coding agents, and context engineering.
Read the essaysProject
Airic
Markdown-configured agent kernel for your workspace via ACP — modes, processes, and tool guidance in editable `.airic/` specs.
Visit AiricAgents
Skills
Installable CMP agent skills — a pre-task planner and a post-change design reflection loop.
Browse skillsRecent Essays
View all- Hermes Cut 34% of Its Code. Did Its Context Footprint Shrink?
A natural experiment on Nous Research's Hermes Agent refactor: measuring whether a 34% code reduction actually shrank the context required to reason about its behavior, and why code decomposition is not context decomposition.
- Will Your Coding Agent Quietly Make Business Decisions for You?
When policy and mechanism sit mixed together, an agent can rewrite business rules while it rewrites implementation — and you may never see it. Clean Architecture Planner splits the two before any code is written.
- Post-Change Design Reflection: Making Every Task Leave the Codebase Cleaner
How a CMP practice turns the context an agent just acquired into design feedback—syncing docs, flagging duplication, and leaving ownership decisions for humans—so each change leaves the codebase easier to modify.
- What's Left for Programmers After AI Takes Over the Code?
Code generation by AI is no longer a prediction—it's observable reality. But the work of solving real-world problems with software isn't going away. A framework for understanding what remains human territory: value judgments, problem definition, and the future of the software practitioner.
Written by Leric Zhang. Find more on GitHub.