Major Developer Tooling Updates from Figma Config 2026 to Ruby AI Framework

개발 도구 | Thu Jun 25 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 6 sources

Announcements covered developer productivity tools including Figma's AI motion and code layers, the RubyLLM framework, the Node.js toolkit Nub, and NVIDIA NeMo AutoModel.

Analysis

[Figma] unveiled AI motion graphics, code layers, and shader tools at Config 2026 ^[1]^[2]

Code layers enable cloning repositories and editing code directly on the canvas
Prompt-based AI motion graphics and 3D transform generation
WebGPU-based shader effects (dither
pixelate
blur) supported
Figma Weave workflow integration and generative plugin authoring

[Figma] added AI agent collaboration features and custom plugin creation tools ^[1]^[2]

Convert repetitive tasks into team-shareable agent skills
Connect third-party connectors including Notion
Granola
Excel
and GitHub
Create plugins like layout generators and vector tracers from prompts alone
Non-technical users can build AI tools without a development environment

[RubyLLM] released a Ruby framework integrating major AI providers ^[3]

Unifies GPT
Claude
Ollama
and others under a single interface
Supports multi-file analysis including images
video
audio
and PDFs
Define agents with tools and instructions via RubyLLM::Agent
Provides structured output
streaming
and Rails ActiveRecord integration
Only three dependencies: Faraday
Zeitwerk
and Marcel

[Nub] released a Bun-style all-in-one toolkit for Node.js ^[4]

Rust-written stock Node augmentation tool
Build-free execution of TypeScript/JSX/TSX with 2.9x faster startup than tsx
24x faster script execution than pnpm run
19x faster than npx
Built-in .env auto-loading and YAML/TOML/JSONC loader support
Integrated Node version manager (replacing nvm/fnm)

[NVIDIA NeMo AutoModel] released a MoE fine-tuning acceleration library based on Hugging Face Transformers v5 ^[6]

3.4-3.7x training throughput improvement over Transformers v5
29-32% reduction in GPU memory usage
Added Expert Parallelism and DeepEP fused all-to-all dispatch
Maintains compatibility with TransformerEngine kernels and the from_pretrained() API
Supports models from Nemotron 3 Ultra 550B to Qwen3-30B-A3B

[NixOS] conducted experiments to reduce default ISO image size ^[5]

Default NixOS ISO is 458MiB
about 7x larger than the Alpine VM ISO (~66MiB)
Reduced to 384MiB by setting nix.enable = false and documentation.enable = false
Tracked unnecessary dependencies like Boost using nix why-depends
Shared a custom ISO build workflow based on the iso-image.nix module

Major Developer Tooling Updates from Figma Config 2026 to Ruby AI Framework

Analysis

Sources