Vercel CEO Guillermo Rauch: README is handwritten because it is for people
Vercel CEO Guillermo Rauch stated that he is working on a new small project. The README is handwritten because it is for people. The internal documentation uses AI-generated English because it is for machines.
He did not disclose the project name, code repository, or launch time. The distinction is placed solely on the readers: the same set of materials tells stories for people and writes internal documents for machines.
He described this as an experience that can make the world better. Blogs, short posts, and READMEs belong to communication between people. The issue is not whether there are dashes, but rather that this is a moment of storytelling that connects using others' words.
Handwriting is limited to consumer-facing content, not all documentation. The internal structure is assigned to AI-generated English, indicating that generated text is left for machines to read, index, and execute. The layer read by humans is required to retain the author's own sentences.
This post did not include product metrics nor did it state that handwriting could improve retention or revenue. It only rewrites the writing process: external explanations are finalized by humans, while internal explanations are generated by models.
This is a split in the documentation supply chain, not a model release. The buyer is someone who reads the README before deciding whether to use it, while the seller is the machine that digests internal documents to execute tasks. The event is driven by the emergence of two readers within the same project. Funding will not shift due to a single post, but the writing budget will be divided into handwritten entry points and internally generated content. The beneficiaries are the human users who need to be persuaded, while the pressure is on teams that hand over external explanations to models, thus losing the storytelling aspect.
Source: Public information
ABAB AI Insight
When Guillermo Rauch created Socket.io and Next.js, the documentation itself was a distribution channel. Developers first read the README before deciding whether to install. Vercel later turned the deployment of Next.js into a hosting platform, and documentation remained the customer acquisition entry point. This time, he leaves the entry to handwritten content and assigns internal pages to AI-generated English, acknowledging that readers have split into two groups: those to be persuaded and those to be retrieved by machines.
Capital has not been directed towards new models. Costs occur in the writing process: human time is spent on READMEs, blogs, and short posts, while model outputs are placed in internal structures that machines will read. The motivation is that the entry is responsible for conversion, while the internal is responsible for execution. Sentences read by humans must connect, while sentences read by machines must be callable. Using the same set of generated text on both sides will lead to a loss of authorship at the entry point.
A parallel can be drawn to the 1990s open-source projects that wrote READMEs as installation stories and man pages as parameter tables. Stripe later wrote guides for developers as narratives and API references as machine-readable structures. Rauch is in the same split version for machines, not an expansion of the content industry.
Structurally, this is a reconstruction of the industry chain. Documentation has transformed from a single publication into two supply chains. The mechanism is that the readers have changed: humans want stories, while machines want executable internal English. The entry is still priced by humans, while the internal is priced by generation costs. When mixed together, persuasion and execution will both deteriorate simultaneously.
ABAB News · Cognitive Laws
Text for people sells connection, while text for machines sells execution.
Handwritten entry allows for internal generation.
Once readers are split, the same writing style is no longer sufficient.