If you mostly use chat-style AI and you're trying not to fall behind, this is the part worth noticing. The easy mistake is to stare at the feature list, then spend time on the wrong signal. ByteDance's DeerFlow (bytedance / deer-flow) matters here for a more basic reason.
My take is simple: in the agent era, the best docs serve machines first. When you're judging an update, don't start with what it can do. Start with whether it changes your next move.
The public README gives one-line setup paths for three coding agents, meaning AI tools that can follow setup steps for you: Claude Code, Codex, and Cursor. Then it sends them to Install.md to finish setup on their own. That is already a signal. Normal docs assume a human will read, translate, and babysit every step.
The stronger signal is the file shape. Install.md is positioned for coding agents, and the repo treats SKILL.md files like capability modules an agent can load when needed. In plain English, the docs are being written like an interface an AI helper can execute, not just a page a person can skim.
That does not prove DeerFlow wins everything. This is a narrow read based on the current public README and setup guide, not user feedback or a full competitor comparison. But it is enough for one useful filter: projects that make their instructions runnable by agents are probably closer to where the product is going. Share this with the person who keeps forwarding AI launches without saying what actually changed.