Presented by
Good morning, AI enthusiasts. Remember when Anthropic was just a safety-focused lab building models for researchers? It's reportedly prepping its most ambitious week yet, with a new flagship model and a first-ever AI design tool arriving as soon as this week.
The same company that built Claude Code for engineers is now building a visual design platform, seemingly taking direct aim at Google AI Studio and its expanding creative toolset. Has Anthropic quietly decided to become a full-stack AI studio?
In today's recap:
Anthropic's flagship Opus 4.7 and AI design tool inbound
NVIDIA opens AI to quantum computing for the first time
Automate repetitive dev workflows with Claude Code Routines
GPT-5.4 cracks an open Erdos math problem
3 new AI tools, prompts, and more
ANTHROPIC
Anthropic's next flagship and AI design tool are nearly here
Recaply: Anthropic is reportedly preparing Claude Opus 4.7 and a new AI design tool for release as early as this week, marking the company's first direct move into visual design tooling alongside a next-generation model.
Key details:
The design tool is a full-stack app platform similar to Google AI Studio. It lets developers and non-technical teams build and deploy AI-powered apps from one place.
Claude Opus 4.6 already has a 1M token context window. Opus 4.7 follows Anthropic's pattern of a new flagship every two to three months, with developers spotting internal API references this week.
Reporter Stephanie Palazzolo broke the scoop. Both products could ship this week, though Anthropic hasn't officially confirmed either announcement.
The release also includes an updated Claude Code interface for managing multiple projects. Developers get one place to handle complex multi-repo workflows.
Why it matters: There has been talk that Anthropic is primarily a model company, content to let others build the tooling layer. But releasing a design tool alongside Opus 4.7 tells a different story. With Google AI Studio offering free visual app-building and Anthropic reportedly building something similar, the battle for where developers build, not just which model they use, is heating up fast.
PRESENTED BY CLICO
🧠 AI that knows the page you're already on
Recaply: Most AI workflows carry the same hidden tax: copy the page, switch tabs, paste your context, then go back and repeat. Clico is a Chrome extension that lives directly inside your browser, understands the page you're currently viewing, and outputs content exactly where your cursor sits.
With Clico, you get:
Clico It: summon AI anywhere on the page and generate output right at your cursor, no tab switching or copy-pasting required
Multi-tab context: use @ to pull multiple open tabs into a single prompt
Memo It: double-press command for an instant summary of any page or PDF
Select to rewrite: highlight existing text and ask Clico to replace it in place
NVIDIA
NVIDIA unveils first open AI models for quantum computing
Nvidia
Recaply: NVIDIA just unveiled Ising, the world's first family of open AI models built for quantum computing, with two model types covering processor calibration and error correction for researchers and enterprises.
Key details:
Ising comes in two parts. Ising Calibration automates quantum processor setup, cutting calibration from days to hours. Ising Decoding handles real-time error correction with two variants optimized for speed or accuracy.
Ising Decoding is 2.5x faster and 3x more accurate than pyMatching, the current open-source error-correction standard. The quantum computing market is projected to hit $11B by 2030, per analyst firm Resonance.
Jensen Huang said "AI is essential to making quantum computing practical," calling it "the control plane, the operating system of quantum machines." Early adopters include Harvard, Fermilab, IQM Quantum Computers, and the UK National Physical Laboratory.
Ising is available now on GitHub, Hugging Face, and build.nvidia.com. It integrates with NVIDIA's CUDA-Q platform for hybrid quantum-classical computing.
Why it matters: NVIDIA's strategy has always been consistent: build the infrastructure layer that everyone depends on, then supply the hardware that runs it. With Ising, NVIDIA is applying the same playbook to quantum that it used in AI. Open models pull researchers and enterprises onto the platform early, and when quantum hardware scales up, NVIDIA will already be embedded everywhere. It's a long game, and NVIDIA is playing it well.
GUIDES
Automate repetitive dev workflows with Claude Code Routines
Recaply: In this tutorial, you will learn how to set up a Claude Code Routine to automate a repetitive development task, such as nightly issue triage or automated PR review, running on Anthropic's cloud infrastructure without keeping your laptop open.
Step-by-step:
Go to claude.ai/code/routines and click "Create routine." Give it a descriptive name and write a self-contained prompt describing exactly what Claude should do and what success looks like, for example: "Scan all GitHub issues opened since yesterday, apply labels based on the affected module, and post a summary to Slack."
Select one or more GitHub repositories for Claude to work in. Each repo is cloned fresh from the default branch at the start of every run, so Claude always operates on the latest code.
Add a connector such as Slack or Linear from the Connectors section. This allows Claude to post summaries, open tickets, or take actions in external services as part of the routine's output.
Configure your triggers: add a schedule (e.g., every weeknight at 11pm) for recurring jobs, an API endpoint if you want to fire the routine from a CI/CD pipeline, or a GitHub event trigger such as pull_request.opened to run automated code review on every new PR.
Save and monitor the first few runs at claude.ai/code/routines. Review Claude's outputs, leave feedback in the run viewer, and refine your prompt until the routine produces consistent, actionable results.
Pro tip: Combine multiple trigger types on a single routine. A PR review routine can run nightly to clean up open review backlog AND fire automatically on every new pull request with a GitHub trigger, covering both scheduled and event-driven review in one setup.
OPENAI
GPT-5.4 Pro cracks an open Erdos math problem
Recaply: OpenAI's GPT-5.4 Pro just solved Erdos Problem #1196, an open combinatorics challenge, becoming the first AI model to crack a problem from the Erdos prize list by finding a 2011 preprint the problem's own author had missed.
Key details:
GPT-5.4 Pro solved the problem in one shot after 80 minutes of thinking time, surfacing a 2011 preprint that provided a proof shortcut the original problem author did not know existed, then building on it to complete the solution.
On Epoch AI's FrontierMath benchmark, GPT-5.4 Pro scored 50% on Tiers 1-3 and 38% on Tier 4; GPT-5.2 Pro had scored 31% on Tiers 1-3 just three months earlier in January 2026, showing rapid progress.
Jared Duker Lichtman, one of the leading human experts on this cluster of problems, described the result as significant; the solution was independently verified on the Erdos Problems forum on April 13, 2026.
GPT-5.4 Pro was released in March 2026; Epoch AI ran an additional 10 trials on Tier 4 and found a pass@10 score of 38%, with another previously unsolved problem cracked in one of those runs.
Why it matters: AI scoring well on math benchmarks is now expected. AI finding a forgotten 2011 preprint and using it to complete a proof that even the domain expert missed is something different. This isn't performance on a leaderboard. It's evidence that GPT-5.4 can synthesize across the entire published research record in ways that individual experts can't. If that capability scales, the implications for research acceleration go far beyond mathematics.
TOGETHER WITH ACCIO WORK
Accio Work: Your Business, On Autopilot
Run your business effortlessly with Accio Work. Our specialized AI agents handle sourcing, supplier deals, store management, and marketing—all automatically. Backed by Alibaba.com’s vast product network, execution is fast and reliable. Skip the complexity—get results instantly while staying in full control of your growth.
TOOLS
Trending AI Tools
🎨 MAI-Image-2-Efficient - Microsoft's new text-to-image model, 22% faster and 41% cheaper
⚙️ Superblocks 2.0 - AI-generated enterprise internal apps
🎨 Wonder - Design tool where everything created is backed by real code
NEWS
What Matters in AI Right Now?
Anthropic released Claude Code Routines in research preview, letting users configure a prompt, repo, and connectors once and run it on a schedule, via API call, or in response to GitHub events on Anthropic's cloud infrastructure.
OpenAI expanded its Trusted Access for Cyber program with GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for defensive security work, with access available to vetted security vendors, organizations, and researchers via tiered identity verification.
Google introduced Gemini Skills in Chrome, letting users save reusable AI prompt workflows that run across browser tabs with a forward-slash shortcut, and launched a design preview in AI Studio that generates custom app themes in seconds while code is being built.
Meta announced an expanded partnership with Broadcom to co-develop multiple generations of next-gen MTIA custom AI chips, with an initial commitment exceeding 1GW as part of a sustained multi-gigawatt rollout to power AI across Meta's apps at global scale.
Apple privately threatened to remove Grok from its App Store in January after xAI failed to sufficiently stop the chatbot from generating sexualized deepfakes, according to a letter Apple sent to senators obtained by NBC News.
Mintlify raised a $45M Series B at a $500M valuation led by a16z and Salesforce Ventures, bringing total funding to $67M to build AI-native knowledge infrastructure for its 20,000-plus company customer base.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
