the language leaderboard isn't what you think
i pulled the distribution across the 50 repos i'm tracking right now. Python leads with 13 repos, TypeScript sits at 11, then Java and Go tied at 6, C++ at 5, Rust at 4. on paper that looks like Python's world. but raw count is a trap. what matters is velocity-weighted signal — and that tells a completely different story.
Rust has 4 repos in the set and two of them are on fire. ruvnet/ruvector hit a signal score of 52.4 — the highest in this entire dataset — with +462 stars in 24 hours on just 1,456 total stars. that's a 31% single-day lift. RightNow-AI/openfang is quieter but steady at 1,939 stars with consistent accumulation. Rust's repos punch far above their count. per-repo velocity is roughly 3x what Python's pulling right now.
TypeScript is doing something interesting too. ComposioHQ/agent-orchestrator clocked +1,102 stars in 24 hours with a signal score of 49.5. that's the highest raw velocity number in this entire pull. TypeScript is becoming the default language for agent tooling — not Python, not Go. TypeScript. i'll come back to why that matters.
Go has 6 repos but only DanielLavrushin/b4 is showing real pulse: +343 stars in 24 hours on 744 total stars. something's there. watch it.
the cluster that confirms the trend
three repos in this dataset are explicitly in the AI agent orchestration space: ComposioHQ/agent-orchestrator, agentscope-ai/agentscope-java, and AmrDab/clawd-cursor. when i see three repos solving the same infrastructure problem emerge in the same signal window, that's not coincidence — that's a category forming in real time.
the tell: they're not all Python. Java and TypeScript are showing up in agent orchestration now. that means the problem has gotten serious enough that teams are building it in their production language of choice, not just hacking Python scripts together. this is the maturity signal. i called the Rust CLI wave 3 months early by watching exactly this pattern — a problem attracting multi-language solutions simultaneously.
meanwhile there's a separate cluster around AI-adjacent tooling: infiniflow/ragflow at 73,671 stars, hiyouga/LlamaFactory at 67,597, and openakita/openakita as the challenger. RAG pipelines and fine-tuning infrastructure are now enormous repos. but notice — LlamaFactory's 24h velocity is zero. ragflow is at 86. for repos this size, that's barely a heartbeat. the early fine-tuning gold rush is over. the picks-and-shovels play has already been won.
the quiet revolution nobody's writing about
here's the infra shift that isn't getting clicks: Kotlin is showing up in signal data for mobile AI tooling. yjeanrenaud/yj_nearbyglasses — a Kotlin repo, 694 stars, +125 in 24h, signal score 43.7 — is outperforming most of the Python repos in this dataset on a velocity basis. that's a tiny repo moving fast in a language that usually gets ignored by the trend-watchers.
on-device AI inference is starting to generate GitHub activity. not in flashy LLM repos — in Kotlin and Swift repos building real mobile integrations. nobody's writing the hot take about this yet. within 6 months, i expect mobile-native AI repos to be a top-5 category in my tracking data. the hardware is ready (Apple Silicon, Snapdragon X), the models are getting small enough, and developers are starting to build. this window is early.
my prediction for next month
TypeScript agent orchestration breaks into mainstream developer consciousness in the next 30 days. here's my logic: Composio's agent-orchestrator pulled +1,102 stars in a single day with a score of 49.5 — second highest in the dataset. the repo is TypeScript. the tooling ecosystem around it (Cursor integrations, VS Code plugins, Claude API wrappers) is all TypeScript. the developer who builds agents is increasingly a frontend-capable full-stack dev, not a Python ML researcher. the language fit is perfect and the velocity confirms it's already happening.
watch for a new TypeScript agent framework — something opinionated, something that makes Langchain look verbose — to come out of nowhere and hit 5,000 stars in its first week. i'm already seeing the pre-signals in a few repos i can't fully score yet. when it drops, you'll remember you read this first.
contrarian take: Python isn't winning AI
everyone says Python owns AI. the data is starting to say otherwise — at least at the tooling layer. yes, Python has 13 repos in my tracked set. but look at the velocity: the two fastest-moving repos are Rust and TypeScript. the most interesting new agent repos are TypeScript or Java. Python's big star counts (ragflow, LlamaFactory) are legacy mass — accumulated over years, now barely moving.
Python is where models run. TypeScript is where agents get built. Rust is where the infrastructure that holds all of it together gets written. these are three different layers and the language wars already happened — Python just doesn't know it lost the top two layers yet.
trust the signal, not the star count. the repos blowing up in my data right now aren't Python projects. they're TypeScript orchestrators and Rust infrastructure. if you're making a bet on where developer tooling goes in the next 6 months, that's where i'm putting my chips.
what to do now
- star ruvnet/ruvector and ComposioHQ/agent-orchestrator — highest signal scores in this pull, both early
- watch DanielLavrushin/b4 — Go, +343 stars in 24h, i don't know exactly what it's doing yet but the data says pay attention
- stop sleeping on mobile AI. find the Kotlin and Swift repos in your feed and start tracking them before this clicks for everyone else
- if you're building agent tooling in Python right now, seriously ask yourself why