Skip to main content
Back

Claude runs full UI testing + debugging from single CLI prompt

Alibaba adds voice-to-app coding, OpenAI Codex now available in Claude Code, Google's Agent Kit gets Java support.

March 31, 2026 4 min read

Hey,

Your daily briefing is ready. You can finally take a break from the AI firehose.

Our algos spent the night splitting signal from noise and pulled the top news, models, papers, and repos.

Here’s the must-read:

Together with:
Sponsor Logo
Summary

Read time: 4 min 34 sec

Top News

▸ Anthropic lets Claude run build, UI testing, and debugging in one prompt

Paragon

▸ Build a RAG-enabled AI agent using production-ready integration patterns

Top News

▸ Alibaba adds audio-visual coding to generate apps from voice and video

Neo4j

▸ Join 40+ sessions to learn how to build smarter multi-agent systems

Top News

▸ Microsoft launches Critique for review and Council for multi-model comparison

Signals

▸ OpenAI enables Codex reviews directly inside Claude Code with plugin
▸ Nous Research ships Hermes Agent update with autonomous video generation
▸ Z.ai launches AutoClaw to run AI agents locally without API keys
▸ Stanford study publishes Meta-Harness to improve LLM systems using full history
▸ Google integrates Java support to Agent Development Kit with new features
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Top News
Anthropic adds computer use to Claude Code, enabling full UI testing and debugging directly from the CLI
60,482 Likes
Grok 4 Fast Benchmark

This update removes the gap between writing code and manually testing it by letting Claude interact with the UI itself. In one prompt, it can build, run, test, and fix an app while observing the screen.

What it does

This feature lets Claude control your computer like a user would.

  • Opens apps, clicks buttons, types inputs, and reads results from the screen
  • Runs UI-based workflows that do not expose APIs or command-line interfaces
  • Uses screenshots to understand outputs instead of relying only on logs
  • Handles GUI tasks that normally require switching out of the terminal

Key features

These capabilities focus on end-to-end development and testing workflows.

  • Builds apps, launches them, and verifies flows without external test frameworks
  • Runs onboarding or user flows in Electron and native macOS apps
  • Reproduces visual bugs like layout issues by resizing and interacting with UI
  • Fixes code and validates changes using screenshots in the same session

How to use it

You enable and run it inside the CLI with minimal setup.

  • Install Claude Code version 2.1.85 or later on macOS systems
  • Enable computer use using the /mcp command in an interactive session
  • Available in research preview on Pro and Max plans only
Enable /mcp
Presented by Paragon
Build a RAG-enabled AI agent in minutes
alpha_signal_image_1

Explore the best practices for building production-ready, integration-enabled RAG and AI agents with guidance from Paragon.

In this series, you'll learn how to:

  • Ingest third-party data from Google Drive, Slack, and more

  • Enforce each source's access controls throughout your RAG pipeline

  • Trigger your AI agents to take action across your users' favorite apps

Includes a written guide, video tutorials, and sample code.

Get the guide
partner with us
Top News
Alibaba ships Qwen3.5-Omni, surpassing Gemini 3.1 Pro on audio reasoning and speech benchmarks
5,427 Stars
Grok 4 Fast Benchmark

Alibaba launches Qwen3.5-Omni, a model that handles text, images, audio, and video in one system. You send mixed inputs and receive text or speech outputs through the same interface.

The key capability is audio-visual coding, where you describe an idea with voice or camera and the model generates working apps or games.

Key Features 

Core capabilities focus on multimodal reasoning and real-time interaction.

  • Surpasses Gemini 3.1 Pro on audio reasoning and speech benchmarks
  • Supports 256K token context across combined multimodal inputs in one request
  • Processes 10+ hours audio and 400 seconds 720p video without splitting inputs
  • Trained on 100M+ hours of multimodal data across speech, video, and text

You can choose between three variants. Plus targets maximum capability for reasoning and long-context workloads. Flash targets lower latency for real-time systems. Light targets smaller deployments with reduced cost.

Build with omni
Presented by Neo4j
How to make your multi-agent systems smarter
alpha_signal_image_2

Join NODES AI to learn how developers build AI systems that reason better with improved context layers.

What to expect:

  • 40+ technical sessions from community experts

  • Context Graph panel discussion with leaders in AI

  • Deep dives into GraphRAG, Agentic Systems, and Memory

  • Live sessions with Q&A

Register here
partner with us
Top News
Microsoft enables Copilot to use Critique for reviewing outputs and Council for comparing multiple model responses
6,249 Likes
Grok 4 Fast Benchmark

Microsoft adds Critique and Council to Microsoft 365 Copilot Researcher, enabling multi-model research workflows. The system combines models from OpenAI and Anthropic to generate and evaluate reports together.

The key change is a split between generation and review, where separate models handle drafting and validation.

Critique: dual-model generation and evaluation

Critique assigns distinct roles to two models in the research pipeline.

  • One model plans tasks, retrieves sources, and produces the initial research draft
  • Second model reviews output for source reliability, completeness, and evidence grounding
  • Uses rubric-based evaluation aligned with structured academic review processes
  • Enforces strict citation checks to ensure every key claim links to sources
  • Achieves +7.0 DRACO score improvement across 100 evaluated research tasks

Council: parallel model comparison

Council runs multiple models simultaneously on the same input prompt.

  • Generates independent reports from each model with separate reasoning and citations
  • Uses a judge model to compare outputs and summarize agreement and divergence
  • Highlights differences in framing, interpretation, and evidence selection across models
  • Surfaces unique insights contributed by each model in the final comparison layer
  • Exposes how models differ on identical research tasks without merging outputs
Test in copilot
Signals
1 OpenAI allows Codex to act as second agent in Claude Code using open-source plugin integration 4,428 Likes
2 Nous Research updates Hermes Agent with video generation and multi-instance agent workflows without human intervention 4,881 Likes
3 Z.ai introduces AutoClaw as a local AI assistant that breaks down and executes tasks 2,819 Likes
4 Stanford study presents Meta-Harness, using coding agents with full history access to optimize LLM harnesses end-to-end 1,423 Likes
5 Google expands Agent Development Kit to Java with context control, memory services, and tool integrations 1,493 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us