How to Work and Compound with AI

2026-05-03 · Source: Eugene Yan · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, long

Summary

This article outlines a structured approach for effective and compounding work with AI, focusing on establishing robust workflows and improving AI systems over time. It details methods for providing AI models with comprehensive context, such as organizing code in `~/src` and knowledge work in `~/vault`, and connecting models to organizational knowledge via `INDEX.md` files. The author emphasizes configuring AI behavior and preferences using `CLAUDE.md` files, which act as behavioral contracts, and creating "skills" for frequently repeated tasks. The framework also covers verification strategies, including shifting error detection left and enabling models to self-verify, alongside scaling work through delegation of larger tasks and parallel session management. Finally, it stresses closing the loop by working in the open and mining session transcripts for continuous configuration updates.

Key takeaway

For AI Engineers and MLOps professionals aiming to maximize AI productivity and system reliability, implement a structured approach to AI collaboration. Focus on establishing clear context, encoding behavioral preferences in configuration files like `CLAUDE.md`, and building self-verification loops. This strategy allows for effective delegation of larger tasks and continuous improvement of AI workflows, ensuring that each interaction compounds knowledge and reduces future errors. Regularly review session transcripts to identify and address gaps in AI instructions and refine your custom skills.

Key insights

Effective AI collaboration requires structured context, configurable preferences, robust verification, and continuous feedback loops.

Principles

• Provide good context to AI models.
• Encode preferences as configuration.
• Make verification easy and early.

Method

Organize context in structured directories, define AI behavior via `CLAUDE.md` and "skills," enable self-verification, delegate larger tasks, and refine configurations by analyzing session transcripts.

In practice

• Use `CLAUDE.md` for AI behavioral contracts.
• Create "skills" for tasks done weekly.
• Mine transcripts to refine AI configurations.

Topics

• AI Workflow
• Context Management
• AI Configuration
• Model Verification
• Task Delegation

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• How to Talk Like an AI Expert — True AI expertise involves understanding integration challenges and maintaining tool-agnostic strategies, not just raw intelligence.
• The Model Wasn’t the Bottleneck. The Configuration Was. — Usable AI capability stems from configuration and human oversight, not just the model's inherent strength.
• Stop Prompting Claude. Start Designing Autonomous Loops. — The core leverage in AI now comes from architecting self-correcting, recursive systems, not single prompts.
• Agentic AI That Actually Works — Enterprise AI success requires embedding agentic AI into governed workflows for controlled, scalable, and auditable execution.
• Claude Opus 4.8: A Smarter Model in the Right Direction — AI model differentiation is shifting from raw intelligence to reliability, cost-efficiency, and graceful failure in production.
• Building AI Agents for Real-World Problems & Workflows — Real-world AI agents thrive as coordination layers, integrating with existing systems and human oversight, not as isolated decision-makers.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Eugene Yan.