Install Claude Code on Windows and fix every error
One PowerShell command installs Claude Code on Windows. Here's the exact fix for every install error: PATH, wrong shell, Git Bash, WSL and more.
ES
Everything we have published
The library
Filter by topic, or hit ⌘K to search titles, decks and tags.
Format
Topic
Seven AI app builders, sorted by what you own at the end
Five of these seven builders will hand you the code. Whether it runs anywhere else depends on the database, the logins and the secrets.
ES
In the works
How we test coding agents
Nine tasks, one deliberately broken repo, and a scoring rule that allows ties. The whole kit is public so you can rerun it and disagree with us.
ES
In the works
Claude Code vs. Codex: which agent actually finishes the job?
We gave both CLI agents the same nine tasks in the same messy repo and scored every round. The gap is not where the marketing says it is.
ES
Hooks: make Claude Code follow your rules automatically
A prompt is a suggestion. A hook is a guarantee. Wire up formatting, guardrails and notifications in about twenty minutes.
ES
In the works
How agentic commerce actually works
Checkout was built for humans with thumbs. Here is what changes when the shopper is an agent with an API key and a budget.
ES
In the works
Build your first MCP server in 30 minutes
Model Context Protocol without the spec-reading. One server, three tools, and a working connection to a real client.
ES
In the works
Cursor vs. Windsurf: the editor war nobody is winning
Two AI-native editors, two philosophies about how much rope to give the model. We used both as a daily driver for a month.
ES
In the works
You are not running out of context. You are wasting it.
Bigger context windows did not fix retrieval. A field guide to what agents actually keep, drop, and hallucinate.
ES
In the works
Subagents in practice: when parallel beats sequential
Fan-out is seductive and usually wrong. Here are the four shapes of work where it genuinely pays, with the configs.
ES
In the works
Teardown: what you are really paying for per agent run
We instrumented 400 real agent sessions and traced every dollar. Cache hits matter more than model choice.
ES
In the works
Writing a CLAUDE.md your agent will actually obey
Most instruction files are wish lists. The ones that work are short, testable, and written like a lint config.
ES
The newsletter
One teardown a week. No launch coverage.
We only send when we have actually run the thing. Unsubscribe in one click.