One-Shot Approach

A technique where an AI agent attempts to generate a complete solution (e.g., full application or complex feature) in a single prompt. Prone to failure due to context window limitations in large models.

Problem

• Context Window Constraints: Models like Claude have fixed context lengths (e.g., 200k tokens), making it impossible to process large codebases or complex features in one interaction.
• Result: Incomplete or inaccurate outputs when attempting “one-shot” generation for non-trivial tasks.

Solution: Incremental Workflow

• Developed by Anthropic and adapted by the video creator to overcome context limitations.
• Breaks tasks into iterative, manageable steps instead of monolithic prompts.
• Enables long-running agents to maintain context through progressive refinement.

Related Concepts

• context-window
• AI Coding Agent
• Long-Running Tasks

2026 04 14 Fixing long running Claude code sessions