I spent the morning going through the 2025 Juejin GitHub trending recaps back to back — April, May, June, and October — and the thing that finally crystallized for me is that one category of project has been on the trending list almost every single month of 2025 and the Juejin AI IDE roundups have barely mentioned it once. Agent-S, Graphiti, supermemory, mem0, FastMCP, claude-cookbooks, TradingAgents-CN, the llm-course repo, the leaked-system-prompts repo. Every one of those is an agent-runtime or memory-layer or prompt-engineering project, and almost none of them appear in the same Juejin posts that score Cursor at the top of the AI IDE list and ChatGPT at the top of the pricing guide. I would not have written that sentence six months ago, and I want to put it down somewhere I can find it.
The piece that pushed me over the edge was reading the April, May, June, and October GitHub trending recaps in sequence and noticing that the same handful of agent-runtime repos kept showing up month after month. April had Agent-S for GUI automation, Graphiti for memory graphs, FastMCP for plugin frameworks, and claude-cookbooks as the Anthropic reference. May had mem0 for persistent recall, supermemory for cross-tool context, and AgenticSeek as a local-first privacy alternative. June had Anthropic Cookbook again and KiloCode as a VS Code agent extension. October came back with supermemory and Agent-S again plus claude-cookbooks still. To be fair the roundups each had ten projects and only four or five of them were agent-runtime pieces, but the fact that Agent-S and claude-cookbooks and the memory-layer repos kept resurfacing across four separate monthly recaps is the part that has been rattling around in my head all morning. That is not a one-month story, that is the story of 2025.
The meta-pattern I want to put down before I forget it is that the agent-runtime category has been the dominant GitHub trending story of 2025, and the Juejin AI IDE and pricing roundups have been telling a different story entirely. The December 2025 IDE ranking scored Tencent CodeBuddy at 9.6, Sourcegraph Cody at 8.2, Codeium at 7.8, Tabnine at 7.6, JetBrains AI Assistant at 7.4, and the rest, with the implicit message that the AI IDE race is between autocomplete-with-chat tools that have been around since 2023. The pricing guide from the same week anchored on the twenty-dollar monthly tier and talked about Cursor and ChatGPT Plus and Claude Pro and GitHub Copilot as if those four were the only game in town. To be fair all four of those tools are real products that I use every week, and I am taking the IDE scores with a grain of salt because the post that gave CodeBuddy the 9.6 was transparently written by an advocate. But the shape of the gap is the part that matters. The GitHub trending recaps are tracking where the engineers are actually pushing code, and the IDE and pricing roundups are tracking where the marketing dollars are being spent, and those are two different markets.
Honestly I think the practical takeaway here is that if you are picking AI tooling in 2026 based on the Chinese-language IDE and pricing roundups alone, you are reading two documents that are six to nine months behind the GitHub trending curve. Cursor and Claude Code are still my daily drivers and ChatGPT Plus is still where I land for everything else, so I am not telling anyone to abandon those. But the projects I am now bookmarking and returning to are Agent-S for the agent-runtime pattern, Graphiti and mem0 and supermemory for the memory-layer pattern, FastMCP for the plugin framework pattern, and claude-cookbooks for the reference implementation. I have not stress-tested Agent-S or Graphiti the way I have with Cursor and Claude Code, so I want to actually ship something on top of them before I oversell or undersell them, but the fact that four separate monthly GitHub trending recaps put these projects on the list and the IDE roundups never did tells me the roundup format itself is scoring the wrong axis, and that is the gap I want to call out before more people make tooling decisions based on it.
I will reassess in three months. The last time I said that I was mostly on Cursor and Claude Code for coding and ChatGPT for everything else, which is still roughly where I land. What has changed is that I now treat the monthly GitHub trending recaps as the source of truth for what to actually build with, and the Juejin IDE and pricing roundups as a lagging indicator for the enterprise procurement market specifically, and I think that split is going to age well. Give it six months and I expect the Chinese-language IDE roundups to start scoring Agent-S and FastMCP and the memory-layer repos alongside the autocomplete tools, and when that happens I will know the roundup format has finally caught up with what the engineers have been shipping on GitHub all year.
Top comments (0)