follow-builders: An AI Agent Skill That Tracks Top AI Builders and Delivers Digests

What follow-builders Does and Who It Is For

The repository's stated philosophy is "follow people who build products and have original opinions, not influencers who regurgitate information." Follow-builders is an agent skill, meaning it runs inside an existing AI agent (Claude Code or OpenClaw) rather than as a standalone application. When installed, it fetches a centrally maintained feed of curated content and uses the agent's language model to produce a summarized digest.

The target user is a developer or researcher who wants to stay current on what working AI practitioners are building and thinking, without monitoring X, YouTube, and blog feeds manually. The digest covers a fixed source list: 26 specific AI builders on X, six AI-focused podcasts on YouTube, and two official blogs (Anthropic Engineering and the Claude Blog). It does not aggregate the broader AI media ecosystem.

The Source List and What Gets Summarized

The 26 X accounts include a mix of researchers, founders, and product managers actively building AI products. The README lists names such as Andrej Karpathy, Amjad Masad, Guillermo Rauch, Sam Altman, Dan Shipper, and Amanda Askell, among others. The six podcasts are Latent Space, Training Data, No Priors, Unsupervised Learning, The MAD Podcast with Matt Turck, and AI & I by Every, all accessed as YouTube playlists or channels.

The two official blog sources are Anthropic Engineering (technical posts from the Anthropic team) and the Claude Blog (product announcements and updates). These are fetched as full articles via web scraping. X posts are fetched via the official API. YouTube transcripts are retrieved via Supadata. All of this collection happens centrally: the repository maintainer runs the feed update daily, and your agent makes a single HTTP request to fetch the result. This design means no API keys are required from the user's side.

The source list is curated centrally and updated automatically. Individual installations receive new sources without any local configuration change.

Installing and Setting Up follow-builders

The skill installs into your agent's skills directory. For Claude Code:

git clone https://github.com/zarazhangrui/follow-builders.git ~/.claude/skills/follow-builders
cd ~/.claude/skills/follow-builders/scripts && npm install

For OpenClaw:

git clone https://github.com/zarazhangrui/follow-builders.git ~/skills/follow-builders
cd ~/skills/follow-builders/scripts && npm install

After installation, invoke the skill in your agent by saying "set up follow builders" or invoking /follow-builders. The agent walks through configuration conversationally and asks how often you want digests (daily or weekly), your preferred time, what language you want (English, Chinese, or bilingual), and how you want delivery (Telegram, email, or in-chat). Your first digest is generated immediately after setup. The only prerequisite beyond the agent itself is an internet connection.

Customizing Summary Behavior Through Prompt Files

The prompts/ folder in the repository contains plain-English instruction files that control how each content type is summarized. The files are:

- summarize-podcast.md: controls how podcast episodes are summarized - summarize-tweets.md: controls how X posts are summarized - summarize-blogs.md: controls how blog articles are summarized - digest-intro.md: controls the overall digest format and tone - translate.md: controls how English content is translated to Chinese

Changes to these files take effect on the next digest generation. The README describes two ways to modify them: by telling the agent what you want in plain English ("Make summaries more concise," "Focus on actionable insights," "Use a more casual tone") and having the agent update the files, or by editing the files directly as a power user. Both approaches work because the agent reads the current file content before generating each digest.

Delivery preferences are also adjustable through conversation after setup. Saying "Switch to weekly digests on Monday mornings" or "Change language to Chinese" updates the configuration without touching any files manually.

Privacy Model and What Gets Stored Where

The privacy behavior is explicit in the README. No API keys from the user are sent to any external service; all content is fetched and processed centrally by the maintainer's infrastructure before your agent retrieves it. If you configure Telegram or email delivery, those credentials are stored locally in ~/.follow-builders/.env on your machine and are not transmitted to the central feed server.

The skill only reads publicly available content: public blog posts, public YouTube videos, and public X posts. Reading history, configuration, and delivery preferences stay on your machine. This is a meaningful distinction from services that store your reading behavior on a server, though it also means your configuration does not roam to other devices.

Limitations and What follow-builders Is Not

The source list is centrally curated and not customizable by the user without modifying the central feed. If you want to add a specific person or podcast that is not in the default 26 X accounts or six podcasts, that requires either a fork or a change request to the repository maintainer.

The skill depends on the central feed being available and updated. If the maintainer stops running the feed update pipeline, local installations stop receiving new content. There is no local fallback that independently fetches from X, YouTube, or blogs. This is the trade-off for the no-API-key design.

An alternative with overlapping intent is setting up your own RSS reader (such as FreshRSS, a self-hosted feed aggregator) and subscribing to blog feeds and podcast RSS feeds manually. The qualitative difference is that FreshRSS presents raw items without any AI summarization or curation, and it requires setting up your own server and managing subscriptions. Follow-builders trades configurability for a lower setup burden and AI-generated summaries.

The repository has an MIT license, no GitHub releases, and the last push was on 2026-09-27.

Editorial conclusion

Developers who use Claude Code or OpenClaw and want a daily or weekly summary of what working AI researchers, founders, and engineers are saying will find follow-builders practical to set up in under ten minutes. It covers a specific, fixed source list and is not a general RSS aggregator or social media monitor. Anyone who wants to track a custom set of people or sources beyond the 26 X accounts and six podcasts curated here would need to modify the central feed configuration, which is maintained externally and not controlled by the local installation.

Frequently asked questions

Official sources

1. Issues
2. README
3. zarazhangrui/follow-builders on GitHub