How Anthropic uses Claude Code: Agentic Software Engineering at Scale
summary
Claude Code exposes concrete primitives for scaling agentic engineering: the edit tool (file name, old string, new string find‑and‑replace schema) proves extremely reliable in practice, while the context window remains roughly one million token and forces aggressive selection. Plugins like skills, MCP and subagents help but hooks scale best by running external scripts only when relevant. Tool search and agent teams, plus features like slashloop and auto mode, work but MCP/tool descriptions do not scale without curation; Mozilla’s result—more bug fixes in one month than the prior 15—shows real impact.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 10:34Agent in agentic programming
- 26:48Context window constraints
- 33:53Claude Code plugins
- 49:45Scaling with work trees and teams
- 57:19Agentic engineering 2026
related talks
16mBeyond the IDE: The Future of Developing Software With Agents
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
23mAnthropic
27mAgentic Engineering: Working With AI, Not Just Using It
01h 28mAI Coding Agents and how to code them
01h 03mAI Coding Agents and how to code them
49mAI Agents: Collaboration in Real-Life Applications
26mBack to the Code Why CTOs Must Get Technical Again to Solve AI’s Real Pro