Gary Tan’s Claude Code setup: thousands tried it, opinions split

Gary Tan, Executive Director of Y Combinator, published his personal Claude Code configuration on GitHub — and the developer community was not indifferent. Thousands downloaded, studied, and tested his settings, dividing opinions roughly evenly: some were delighted, others subjected the approach to sharp criticism. Y Combinator is one of the world's most influential accelerators, backed by Airbnb, Stripe, OpenAI, and hundreds of other companies. When its CEO publicly shares his working toolkit, it automatically attracts the attention of the professional community. Tan is not a typical venture investor — he actively writes code, openly experiments with AI tools, and documents the process, which makes his configuration particularly interesting as a reference point.

Claude Code is a console assistant from Anthropic that allows you to work with code directly from the terminal through dialogue with a language model. The tool is gaining popularity among developers who want to embed AI into their usual workflow without switching between applications. Configuration determines the behavior of the assistant: its style, constraints, context, and automations. This is precisely why someone else's config is not just a set of settings, but a snapshot of someone's philosophy of working with code.

The reaction in comments and threads turned out to be mixed. Some developers perceived the publication as a valuable template and began adapting the approach to their tasks. Others criticized it for being excessive or not matching the real scenarios of an ordinary engineer's work. There were also remarks that the configuration is optimized for the specifics of a person who evaluates startups, rather than writes production code for hours a day.

A separate topic for discussion was that language models themselves — Claude, ChatGPT, and Gemini — shared their positions on Tan's settings. This gave the discussion an unexpected and instructive character: AI systems were evaluating exactly how people configure them. Such a turn reflects a new reality — users increasingly treat models not as a tool with fixed behavior, but as a co-author whose opinion you can ask about your own workflow.

In a broader context, the story of Tan's configuration illustrates a shift in development culture. Previously, programmers shared dotfiles, snippets, and vim configs. Now system prompts, CLAUDE.md files, and AI agent settings have been added to that. GitHub is gradually becoming a space where developers showcase not only code but also their relationships with artificial intelligence.

The polarity of reactions is easy to explain: the market for AI tools for development is not yet settled, and each specialist develops their own workflow. A configuration ideal for one context may be useless or even hindering in another. This is not a weakness of Tan's approach — it is simply a sign that tools like Claude Code are in a phase of active exploration, where best practices do not yet exist, but there are live experiments.

The viral success of the publication speaks to the main point: developers are actively searching for guidance on how to properly integrate AI into code work, and are ready to learn from others' experience — even if they then argue about the details. In this sense, any config that makes the community think and discuss has already fulfilled its function.