SGLang structured LLM execution and serving framework
Fast execution framework for complex multi-turn LLM agent pipelines using RadixAttention for automatic KV-cache reuse and structured decoding.
Image: GitHub
The weekly · Week 12 · 16 Mar – 22 Mar 2026
AI coding agents: what changed
From structured inference to autonomous workflows, this week showcases the SGLang execution engine, Roo Code and CodeCompanion editor assistants, and the STORM deep-research agent.
4 picks
Roo Code multi-role AI coding agent for VS Code
A community-driven VS Code coding agent with custom system modes, MCP tool support, sound cues, and granular file diff tracking.
Image: GitHub
CodeCompanion AI companion and agent for Neovim
An extensible AI companion for Neovim that integrates LLMs for inline code transformations, code review chats, and custom agent workflows.
Image: GitHub
STORM automated deep research and report generation agent
An LLM-powered multi-perspective research system that searches the web, conducts simulated perspective interviews, and generates structured long-form technical syntheses.
Image: GitHub
Picked from public sources (GitHub, Hacker News, DEV, Lobsters, Reddit, Product Hunt) by our daily scout. Each summary uses only what the source itself says; GitHub stars are as checked on the date shown.