The star economy decoupled from the dependency economy

What happened

GitHub's trending page this week is a snapshot of a quiet but important inversion. The three repos at the top of the charts are freeCodeCamp/freeCodeCamp at 437.9k stars, kamranahmedse/developer-roadmap at 350.5k stars, and openclaw/openclaw — a personality-forward AI assistant marketed as "the lobster way" — sitting at 283.1k stars. None of these projects ship as a dependency in anyone's `package.json`. None of them get `npm install`ed into a production service. Two of them are curricula. The third is an agent shell with a mascot.

For roughly fifteen years, GitHub's trending list functioned as a leading indicator of what developers were about to put into production: React, Kubernetes, TensorFlow, Rust, Next.js, LangChain — stars-per-week tracked real dependency graphs. That correlation is breaking down. The current leaderboard reflects two demand signals that have nothing to do with shipping runtime code: people learning how to become developers, and people trying to augment the developers they already are.

freeCodeCamp has held a top-five slot on GitHub for most of the last decade. developer-roadmap has been climbing steadily since 2017. openclaw is the new entrant — it crossed a quarter-million stars in under a year, which is framework-adoption velocity applied to what is functionally a chat wrapper with opinions.

Why it matters

The conventional reading of a GitHub star is "someone might use this." That reading is now wrong about the top of the chart.

Consider what each of these repos actually is. freeCodeCamp is a nonprofit's curriculum repository — HTML, markdown, and challenge definitions for a learn-to-code platform. You star it because you used the site, not because you intend to `import` it. developer-roadmap is a collection of SVGs and annotated trees showing what a frontend engineer, a DevOps engineer, or an ML engineer should learn in what order. It is a career map, not a library. openclaw is a personality-driven agent — the README is heavier on its mascot and tone than on its plugin API. None of these repos participate in the dependency economy. All three dominate the attention economy.

This split matters because the industry still uses stars as a proxy for adoption when making purchasing, hiring, and architectural decisions. VCs benchmark funding rounds on star velocity. Engineering managers cite star counts in RFC documents. Vendors buy ads promising "the #1 open source tool for X." If the top of the chart is now curricula and charismatic assistants, the signal-to-noise ratio for "should I bet my stack on this" has collapsed.

freeCodeCamp's persistence at the top tells one story: the pipeline of new developers has not slowed down, and the canonical free resource for that pipeline is still a GitHub repo. Layoffs, AI panic, and "learn to code is dead" thinkpieces have not dented the underlying demand for entry-level programming education. If anything, developer-roadmap's rise alongside it suggests the audience has gotten more sophisticated — people aren't just learning to code, they're trying to understand what specialty to commit to in a rapidly fragmenting job market.

openclaw's rise is the more interesting data point. A personality-first AI assistant hitting 283k stars is evidence that developers now pick agents the way they used to pick text editors: based on vibe, tone, and tribal affiliation, not feature matrices. The "lobster way" framing is not incidental. It's the whole pitch. Compare this to how Vim vs. Emacs, or tabs vs. spaces, or React vs. Vue debates worked — the functional differences were smaller than the identity differences, and the identity differences drove adoption. AI assistants have entered that same phase faster than any tooling category in memory. Claude Code, Cursor, Aider, Cline, Codex, Zed's agent mode, and now openclaw — they all do variations of the same thing, and users defend their choice with the energy of someone defending their religion.

There's a third signal embedded in all this: the absence of frameworks from the top of the chart. A year ago you'd expect to see Next.js, Bun, Astro, or a rising Rust crate somewhere in the top ten. This week you don't. Framework churn has slowed, or at least its star velocity has. That's partly because the incumbents (React, Vue, Django, Rails, Spring) are entrenched, and partly because AI assistants have absorbed the "try a new thing every weekend" energy that used to power framework-of-the-month cycles. Why learn a new meta-framework when your agent can scaffold any of them on demand?

What this means for your stack

First, stop using GitHub stars as a primary adoption signal for anything below the top of the chart. If the top of the trending page is curricula and agents, the middle of the page is noise, astroturfing, and projects that gamed a Show HN launch. For real adoption signals, look at `npm` download counts, `pip` install counts, Docker Hub pulls, dependency graph data from tools like Libraries.io, and `go.mod` / `Cargo.toml` references across open-source codebases. Those numbers are harder to manipulate and they track what's actually running in production.

Second, if you're evaluating AI coding assistants for your team, treat the category the way you'd treat editor selection: let people pick, but standardize on the interface. openclaw's success suggests users will fight over personality. They will not fight over MCP servers, prompt formats, or tool-calling conventions. Standardize on those. Let individual engineers pick the agent that matches their taste as long as it speaks your team's protocols.

Third, if you're a maintainer of a real library — something that lives in a `package.json` somewhere — the implication is uncomfortable. Your star count is now competing for attention with curricula that millions of beginners use and AI assistants that millions of professionals talk to daily. You will lose that comparison. Don't optimize for it. Optimize for download counts, production references, and whether engineers at companies you respect cite your tool in their postmortems. Those are the dependency-economy signals, and they still mean what they used to mean.

Looking ahead

The next twelve months will probably stretch this decoupling further. Expect GitHub's trending page to become more useful as a cultural barometer — what developers are learning, what identities they're affiliating with, which assistants they're evangelizing — and less useful as a dependency forecast. That's not a bug of the platform; it's what happens when a single metric is asked to do too many jobs for too long. The smart move is to stop asking stars to tell you what to install, and start asking them what your future hires will believe about the industry before they show up on day one. On that question, the current leaderboard is crystal clear: they'll arrive knowing freeCodeCamp's curriculum, following a roadmap someone drew in 2017, and already loyal to an AI assistant with a mascot.