How Claude Code’s New Learning Mode Turns AI into a Coding Coach

Background and Motivation

Recent research from MIT suggests that while AI can boost short‑term productivity, prolonged reliance may lead to "cognitive atrophy" similar to muscle loss, reducing problem‑solving abilities. This concern has sparked debate about AI‑assisted coding tools and their long‑term impact on developers.

Claude Code’s Learning Mode

Anthropic has responded by launching a new "Learning Mode" for Claude Code. In this mode, the assistant deliberately pauses during code generation and inserts TODO(human) markers, requiring the user to complete the task before the model proceeds. This "learn‑by‑doing" approach is designed to keep programmers actively engaged and reinforce understanding.

Output Styles

Claude Code now supports two built‑in output styles:

Explanatory : The model walks the user through its reasoning step‑by‑step, providing a lightweight teaching experience.

Learning : The model and user alternate writing code, with the assistant stopping at critical points for the user to fill in the gaps.

Both styles modify Claude’s system prompt, disabling default code‑generation shortcuts (e.g., concise answers, automatic testing) and inserting custom instructions that encourage active participation, manage cognitive load, foster metacognition, spark curiosity, and deliver supportive feedback.

How to Activate Styles

Users can switch styles with the slash command /output-style. For example, /output-style:explanatory or /output-style:learning. Custom styles can be created via /output-style:new, which saves a Markdown file in ~/.claude/output-styles (or a project‑level .claude/output-styles directory) containing a name, description, and custom instructions.

---
name: My Custom Style
description: A brief description of what this style does
---
# Custom Style Instructions
You are an interactive CLI tool that helps users with software engineering tasks.
[Your custom instructions here...]
## Specific Behaviors
[Define how the assistant should behave in this style...]

Comparison with Other AI Learning Modes

The design mirrors OpenAI’s “Study Mode” for ChatGPT, which also inserts pauses and prompts users to reflect before proceeding. However, Claude’s implementation focuses on code‑centric tasks and integrates directly with its output‑style system.

Technical Details

Changing the output style replaces Claude’s default system prompt with a style‑specific prompt, affecting the main reasoning loop but not the underlying model. Output styles can be shared across projects by storing the Markdown files in the user‑level directory.

References

Engadget article on ChatGPT Study Mode

LinkedIn post by Drew Bent announcing Claude’s Learning Mode

Anthropic documentation for Claude Code output styles