Presented by
Good morning, AI enthusiasts. Anthropic just gave Claude Code a pair of hands. It can now open your apps, click through interfaces, and test software it builds, all from the same CLI where it wrote the code.
The new Computer Use feature works as a last-resort tool in Claude Code's toolkit, kicking in only when there's no API, no MCP server, and no Bash command that can reach what needs to be tested. Narrowly scoped, and that might be what makes it actually useful.
In today's recap:
Anthropic's Claude Code gains computer use
Microsoft Copilot runs multiple AI models together
Automate UI testing with Claude Code Computer Use
Mistral raises $830M for European AI infrastructure
4 new AI tools, prompts, and more
ANTHROPIC
Claude Code now tests what it builds
Anthropic/Screenshot
Recaply: Anthropic just rolled out Computer Use in Claude Code, letting it open apps, click through interfaces, and test what it builds from the CLI. It's now in research preview for Pro and Max users.
Key details:
Computer use activates when no other tool can reach the task, covering native apps, simulators, and tools without an API. Developers can skip Playwright configs and test harnesses entirely.
The feature requires Claude Code v2.1.85 or later and a Pro or Max plan, limited to macOS and one active session at a time.
Claude uses a preference cascade before reaching for computer use: MCP server first, then Bash, then Claude in Chrome. Screen control is reserved for things nothing else can reach, according to the official docs.
Live now in research preview on Pro and Max plans on macOS. Team and Enterprise plans aren't included, and it won't work in non-interactive CLI mode.
Why it matters: Computer use feels like the last piece of a coding loop. Claude Code could already write code and run terminal commands, but it couldn't check that a UI actually worked. Now it can open the app, click through every button, and screenshot the result, all in the same conversation where it wrote the code. For developers building native Mac apps or testing without a harness, that's a meaningful change.
PRESENTED BY SUPERHUMAN AI
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning
MICROSOFT
Microsoft's M365 Copilot goes multi-model
Microsoft
Recaply: Microsoft just introduced Critique and Council to M365 Copilot Researcher, using multiple AI models together for deep research. Critique runs Anthropic and OpenAI models in a two-stage review process.
Key details:
Critique splits the work: one model plans and drafts, while a second reviews and refines. Council runs multiple models on the same prompt at once, then a judge model compares where they agree and differ.
Researcher with Critique scores 57.4 on the DRACO benchmark, a 13.88% improvement over Perplexity Deep Research with Claude Opus 4.6, across 100 research tasks in 10 domains.
The biggest gains were in breadth and depth (+3.33), presentation quality (+3.04), and factual accuracy (+2.58). All improvements were statistically significant, according to Microsoft's evaluation.
Critique is now the default in Researcher when Auto is selected. Council is available via the Model Council option. Both are live in Microsoft's Frontier program.
Why it matters: Microsoft's multi-model approach is a strong bet. By running both Anthropic and OpenAI models together, it sidesteps the debate over which AI is best. For enterprise users doing market research, Critique adds a second reviewer before the final report lands. It also signals where enterprise AI is going: not one model, but several working together.
TUTORIAL
Automate UI testing for your apps with Claude Code Computer Use
Recaply: In this tutorial, you will learn how to use Claude Code's Computer Use feature to automate UI testing for your apps, so you can validate interfaces without writing a single test file.
Step-by-step:
Open a Claude Code session and run
/mcp. Enable the computer-use server from the MCP menu and approve the macOS permissions prompt. You need Claude Code v2.1.85 or later and a Pro or Max plan.Open your project in Claude Code. Prompt: "Enable computer use, then build me a [describe your app] macOS or Electron app and run through the full onboarding flow to confirm it works."
Claude writes and compiles the code, then launches the app using computer use. It clicks through each UI element and screenshots every step automatically.
Review Claude's test output in the CLI. It lists what broke, which buttons didn't respond, and which layouts clipped, and proposes fixes in the same conversation.
Prompt: "Apply the fixes, re-run the onboarding flow, and confirm everything passes." Claude iterates until the UI is clean without you touching a test file.
Pro tip: For regression testing after code changes, prompt: "Run the same UI flow from our last session and flag anything that looks different from the previous screenshots." Claude compares runs automatically.
MISTRAL AI
Mistral bets $830M on European AI infrastructure
Recaply: Mistral AI just secured $830M in debt from seven banks to build a data center near Paris with 13,800 NVIDIA Grace Blackwell chips. It's the company's first debt raising.
Key details:
Seven banks financed the deal, including BNP Paribas, Credit Agricole CIB, HSBC, and MUFG. The facility in Bruyeres-le-Chatel will run on NVIDIA Grace Blackwell infrastructure.
The Paris site will hold 13,800 chips. Mistral is targeting 200MW of capacity across Europe by the end of 2027. A second facility in Sweden is also planned.
CEO Arthur Mensch told Reuters: "Scaling our infrastructure in Europe is critical to ensure AI innovation and autonomy remain at the heart of Europe."
The Paris data center is expected to open in Q2 2026. Mistral selected the site in February 2025.
Why it matters: Europe has been building toward a credible AI alternative to the U.S. and China, but infrastructure was always the gap. Mistral's $830M raise is the most concrete step yet, moving from model development into compute ownership. For governments and enterprises worried about data sovereignty, Mistral is now building the physical layer to back its independence pitch. Whether 13,800 chips and 200MW is enough to compete at frontier scale is the open question.
TOOLS
Trending AI Tools
🤖 PokeeClaw - Pokee AI's always-on agent with 1000+ app integrations
⚙️ Super Agent - CREAO's AI agent that creates schedulable workflow applications from plain-language descriptions.
💻 Hyperbox - Always-on cloud workstation for coding agents
🔧 OpenAgents Workspace - Open-source platform for connecting multiple AI agents in shared workspaces
NEWS
What Matters in AI Right Now?
OpenAI launched a Codex plugin for Claude Code, installable via
/plugin marketplace add openai/codex-plugin-cc, with commands for code review, adversarial review, and code rescue.Stanford researchers published Meta-Harness, a system that autonomously improves agent harnesses by reading full execution traces. It reached 46.5% on a hard TerminalBench 2 subset, compared to Claude Code's 28% baseline.
U.S. studios are reportedly bypassing a voluntary delay on Seedance 2.0's global rollout, accessing ByteDance's cinema-quality AI video tool despite pressure from Hollywood industry groups.
Enrichlabs introduced Helena, an autonomous AI marketing agent with 100+ tools covering SEO content, ads, email campaigns, and social media, with no dev setup or API keys required.
Alibaba's Qwen team released Qwen3.5-Omni, a fully omnimodal LLM trained on 100M+ hours of audio-visual data, supporting 113 languages in speech recognition and 36 in speech generation, available via offline and realtime APIs.
XCath Robotics completed the world's first telerobotic stroke surgery, with neurosurgeon Vitor Mendes Pereira operating 120 miles away in Santiago, Panama, while his patient was in Panama City.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
