How /verifier-setup automates PR reviews in one command

#1 📝 Anthropic News

Introducing Claude for Teachers - Anthropic is launching Claude for Teachers, giving verified U.S. K–12 educators free access to premium Claude capabilities plus a Learning Commons connector that maps academic standards and fine‑grained learning progressions in all 50 states, a library of teaching skills co‑developed with Learning Commons, integrations with K‑12 tools (e.g., ASSISTments, Brisk Teaching, Canva Education, TeachFX), and features like Claude Code and Cowork to analyze class data and schedule recurring instructional tasks. The product restricts teacher‑shared data from model training, includes K‑12 privacy terms and a FERPA‑compliant Data Processing Addendum, partners with the American Federation of Teachers, offers an AI Fluency course, publishes an open‑source skills repo, and will pilot in Detroit Public Schools.

Also covered by: @Claude

#2 𝕏

Claude launched Claude for Teachers, offering verified U.S. K–12 educators free access to premium AI capabilities, a library of teaching skills, and direct integration with evidence-based curricula mapped to standards in all 50 states.

Also covered by: @Claude

#3 𝕏

Sam Altman says GPT-5.6 “sol” runs at half the price and roughly twice the token efficiency of Fable, cutting costs to a quarter for equivalent tasks.

#4 𝕏

Jason Zhou launched the `/verifier-setup` skill which, in one run, scaffolds a verifier sub-agent driving the real app, embeds screenshots/videos in every PR, and provides a one-command local or cloud dev stack—so reviews go from manual testing to “watch the 20-second video, ...

#5 𝕏

Philipp Schmid provides a Python snippet for Google’s new Antigravity Agent on the Gemini API, detailing model configuration, prompt structure and sample responses (gist.github.com/85b595886fb5386ba5a270f3ba36b88f). He also links to the official ai.google.

#6 ▶️

Why I switched to Pi...

AI Jason

#7 𝕏

LlamaIndex 🦙 launched Conversational Extract in LlamaParse—a chat-driven tool that auto-generates and tweaks JSON extraction schemas from your uploaded docs or ready-made templates, so you never have to write schemas by hand.

#8 𝕏

Aravind Srinivas open-sourced WANDR, Perplexity’s internal benchmark suite powering its cost- and performance-leading deep and wide research harness.

#9 𝕏

There's An AI For That published a readable side-by-side benchmark on Hyperagent, letting you compare real AI outputs and see which model best matches your writing and design style.

#10 𝕏

NVIDIA AI fine-tuned its zero-shot Cosmos 3 Nano with LoRA, boosting WTS validation accuracy from 54.41% to 87.14% by correctly detecting a missed traffic signal. Using NVIDIA TAO AutoML further raised accuracy to 93.35% in under a day.

#11 𝕏

NVIDIA AI used autoresearch with NeMo RL, NeMo Gym and reusable skills to task a coding agent with building and training Qwen3-VL-2B to count colored stars, boosting accuracy from 25% to 96.9%. The agent even autonomously proposed the next experiment.

#12 𝕏

Santiago unveils Coder Workspaces—a self-hosted dev platform (on-prem, AWS, etc.) that secures team environments, offers a built-in agent harness (or supports Claude Code, Codex, Cursor), plus modular configs and BYOM support for local models.

#13 𝕏

Santiago launched an MCP that embeds Claude into After Effects projects, letting you guide the AI to generate expressions and automate repetitive tasks directly in your AE workflow via a quick demo video.

#14 𝕏

Guillermo Rauch opened the Vercel AI Gateway token-flow dataset, unlocking detailed insights into AI request patterns, token usage, and performance.

#15 𝕏

Guillermo Rauch rolled out AgentMail on Vercel—just run `vercel install agentmail` for zero-signup, automatic setup and unified billing.

#16 𝕏

Aravind Srinivas integrated Wide Research into the Perplexity Agent API, enabling AI agents to tap into expanded research pipelines for richer, data-driven answers.

#17 𝕏

Harrison Chase standardized coding agent tracing in LangSmith, creating uniform telemetry and debugging across agent workflows.

#18 📝 OpenAI News

How to manage AI investments in the agentic era - Guidance on approaching AI investments in a new era of agentic systems, discussing strategies for managing risk and capturing value. The piece is positioned for leaders and investors navigating AI adoption.

#19 𝕏

clem 🤗 – Co-founder & CEO @HuggingFace forecasts that in the next 12 months, 90% of tokens will go to open models. He observes companies experimenting with frontier APIs while running production and heavier workloads on open-source or private models.

#20 𝕏

Logan Kilpatrick advises companies to up their AI ambitions every three months to capitalize on the latest model improvements, or else hand over the capability overhang to competitors.

#21 𝕏

Teresa Torres introduces her “ladder of evidence” framework, contrasting behavioral analytics with story-based user interviews to help PMs distinguish low- from high-quality signals.

#22 𝕏

Mustafa Suleyman proposed a Financial Stability Board for AI with Ian Bremmer in a summer 2023 Foreign Affairs article, outlining a governance framework to monitor and manage systemic AI risks.

#23 𝕏

Mustafa Suleyman proposed an “IPCC for AI” with @ericschmidt in 2023—a global, science-based council to standardize AI risk assessments and steer policy.

#24 𝕏

AI at Meta submitted its advanced reasoning, multimodal AI model to the Asian Physics Olympiad’s theoretical exam and achieved a perfect 30/30 score, tying with the top three student contestants.

#25 𝕏

Philipp Schmid used one Gemini Managed Agents API call to spin up an autonomous financial analyst—linking a remote MCP server and a zarazhangrui/frontend-slides GitHub skill in a Linux sandbox with real-time Yahoo Finance data—to auto-generate a 6-slide executive deck complet...