AI Tool Changelog

Supabase ships MCP server, UI library, Postgres LSP

Gemini 3.5 Flash launches GA at 3x prior cost

Skybridge 1.0 stabilizes MCP app development

Forge lifts 8B models to agent-class reliability

Treat Claude Code as autonomous agent with guardrails

Run local speech pipeline for Reachy Mini robots

Anchor formalizes ERP agent benchmarking with constraint optimization

Next.js fixes Turbopack imports, devtools, benchmarking

Point GitHub Copilot Chat at any OpenAI-compatible API

Smaller models leak privacy under adversarial probing

Single neuron disables safety across model families

Measure prompt cache hits to verify cost savings

DecisionBench measures router fidelity across agentic delegation

Datasette Agent ships conversational SQL interface

GitHub Copilot shifts usage-based billing June 1st

Gemini Omni Flash generates video from multimodal input

Node.js patches nine vulnerabilities across active releases

HealthCraft measures LLM safety collapse under clinical pressure

Gemini 3.5 Flash executes agentic workflows at scale

Composer 2.5 improves long-task execution and collaboration

Cloudflare Sandboxes Execute Claude Managed Agents

Gemini 3.5 Flash executes agentic tasks at scale

Gemini 3.1 Flash-Lite launches at scale pricing

Qwen3.7 Preview ranks top LLM arenas now

Semble indexes codebases, cuts agent token use 98%

Apple Silicon costs three times more than OpenRouter

LLMs stay blank, you absorb the learning

SMCEvolve cuts LLM calls for program search

Claude Code /goal asks too many questions mid-run

Quantization hides bias emergence below perplexity thresholds

Simon Willison launches LLM briefing newsletter

OpenAI models switch endpoints for interleaved reasoning

GPT 5.5 unlocks autonomous loops for tech debt