Back to feed
News Story
APriority76
InfoQ
1 sources

Claude Code Has No 'Magic': A Deep Dive into Context Engineering and Plugin Design

Based on Daisy Hollman's talk at NDC Copenhagen, this article details the design of Anthropic's Claude Code plugins and the context engineering primitives they constitute, as well as how Anthropic operates multi-agent workflows internally. It highlights customization, hook mechanisms, and the shift from feeding information into models to outputting it to users, reflecting the latest progress in coding agents.

SynthePulse Insight · AI deep readingMembers

Claude Code Has No 'Magic': How Context Engineering Is Becoming the New Core of Software Engineering

Version 1 · 1 source

In her NDC Copenhagen talk, Anthropic's Daisy Hollman deconstructs the design philosophy behind Claude Code: the context window is a finite box, customization is knowledge, and context engineering is becoming a primary discipline in software engineering.

  • Context window sizes have plateaued at around 1 million tokens for a year, while model capabilities continue to grow, forcing engineers to be smarter about what they put into the context.
  • Customization is knowledge: after model weights are frozen, almost all customization happens at the text level, and programmers are best at controlling text.
  • Claude Code's editing tools are essentially find-and-replace; models can already handle complex nested JSON perfectly, but the tools remain primitive, leaving huge room for iteration.
Open section navigationThe Context Window: A Finite Box

The Context Window: A Finite Box

Daisy Hollman points out that while model capabilities have exploded over the past year, context window sizes have largely stagnated: the first 1 million token context windows appeared in late 2024, the most advanced models in February 2025 had 1 million tokens, and it's still the same now. This means that as tasks become more complex, engineers must be increasingly smart about what they choose to put into the context.

The context window is the set of tokens the model can see when predicting the next token, and it's the space for customization. You can't throw the entire codebase in; you must follow the zero-overhead principle of 'don't pay for what you don't use.' Every customization competes with the workspace, and the more you pile up front, the less space remains for the actual task.

KV caching imposes hard constraints: to predict the next token, all previous tokens must be consistent, causing some operations to cost 10 times more. Cursor early on tried LRU caching on Cursor Rules but quickly found the cost prohibitive, indicating the problem is more subtle than simple caching.

Free for now

Read the full analysis

4 more sections of analysis, plus the full takeaway

Loading

Credibility boundary

This article is based on Daisy Hollman's talk at NDC Copenhagen, compiled by InfoQ, and is a second-hand retelling. All opinions and descriptions come from the speaker herself and have not been independently verified. The METR chart and Mozilla data are cited in the talk, and specific details have not been verified.

Sources for this version

  1. Claude Code没有"魔法"InfoQ

Primary report

InfoQ

Primary source