The Inference Report

The GitHub trending set reveals a widening divergence between two developer priorities: raw performance and AI integration. Zed's rise reflects genuine frustration with editor latency, built on concrete engineering choices like Rust and a multiplayer architecture inherited from its creators' experience with Atom. That's not viral momentum; it's a tool solving a specific friction point that affects developers every day. Meanwhile, the traditional infrastructure anchors, React, Rails, VS Code, Prometheus, remain gravitational centers, but they're no longer the conversation's center of gravity. The real motion is happening in the discovery set, where the patterns snap into focus: frameworks for building semantic reasoning layers (Auto Ontology), specialized editors for AI workflows (Plate), and agent skill systems (Google Skills, Cloudflare's security audit skill) are gaining real adoption. These aren't replacements for existing tools so much as new layers stacked on top of them.

The agent skills pattern deserves attention. Cloudflare's security audit skill, Google Skills, and the mattpocock/skills repo all point toward a shift in how developers are packaging reusable intelligence, not as libraries or services, but as declarative, composable units that LLM agents can invoke. This sidesteps the traditional framework problem: instead of committing to a monolithic system, you compose capabilities. The discovery repos also show developers investing in fundamentals that frameworks tend to gloss over: the Study-Reinforcement-Learning guide, AlbumentationsX for image augmentation, and the ai-engineer-notebooks collection all suggest people are building from first principles rather than accepting abstraction layers as gospel. Ruff's continued growth as a Rust-based Python tool signals that performance gains from language rewrites are no longer novelty, they're expected. The infrastructure stack itself hasn't shifted, but what developers are building on top of it has. The question isn't whether to use React or Rails anymore; it's what reasoning layer, what agent capabilities, and what on-device AI you're bolting alongside them.

Jack Ridley