Welcome back, Superhuman. OpenAI has made serious progress on that “AI superapp” it’s been hinting at for months, revealing two major product launches yesterday. Plus, Meta shipped something so important that CEO Mark Zuckerberg broke a three-year silence on X to announce it himself.
Today: OpenAI’s new cross-device agent, how to turn your PDFs into a Claude skill, and get the latest prompts and trending social posts.
TODAY IN AI
1. OpenAI launches ChatGPT Work, powered by GPT-5.6: ChatGPT Work is OpenAI’s new agent that can take actions across your computer or phone to get work done, pulling context from other files to match your style. The AI lab claims it can focus on a single project for hours. It’s rolling out now, along with a brand-new desktop app that bundles ChatGPT, Work, and Codex under one roof (free for all plans). Try ChatGPT Work or see how teams use it.
2. Meta debuts first paid model, pivoting from its open-weight strategy: Muse Spark 1.1 is Meta’s new multimodal reasoning model, built for agentic tasks, coding, and computer use — and for the first time, Meta is charging for it. AI chief Alexandr Wang called the pricing “very aggressive and attractive” when compared to other frontier models. The pivot comes as Wall Street pressures Meta to justify its massive AI infrastructure spend.
3. PromptQL relaunches as a multiplayer AI coworker: PromptQL is repositioning itself as "the first AI version of Slack,” giving teams a shared AI agent that operates across a common thread, helps get work done, and can run on any frontier LLM. One key feature is memory: the tool turns every correction into a reusable skill — so it learns and improves from its mistakes over time. The startup has raised $136M to date, and its relaunch video went viral with 2M+ views. Learn more here.
PRESENTED BY VIKTOR
You spend 3 minutes a day learning what AI can do. Viktor spends all day actually doing it. Viktor is the AI employee that lives in your Slack and Microsoft Teams. Give it a goal and it runs the whole job end to end: pulls the data, drafts the report, chases the people who owe you answers, and hands back finished work ready to send.
Not another chatbot you have to prompt. A teammate that ships, connected to 3,000+ tools you already use and backed by a $75M round. No new headcount to hire. Hand off the busywork and get your mornings back.
THE FRIDAY FOUR
In case you missed them: this week’s biggest stories

Made with Midjourney
OpenAI upgraded ChatGPT’s voice and released its new flagship model publicly. The AI lab launched GPT-Live, a new family of voice models that listen and speak at the same time, making chatting with ChatGPT sound much more human-like. It's rolling out now inside ChatGPT. Separately, the lab’s GPT-5.6 model family went public after the US Commerce Department lifted a weeks-long restriction.
Anthropic gave us a look inside Claude’s brain, updated Cowork, and extended Fable. Fresh research revealed that Claude has a hidden thinking space where it reasons through ideas without ever spelling them out in its chain-of-thought (video here). The lab said this mirrors the human brain: similar to how you might chat about one topic while thinking about another. Also, Claude Cowork now works across devices, and access to Claude Fable 5 has been extended through July 12.
Meta's Superintelligence Labs dropped three new models. Muse Image — which is ranked #3 on Arena's text-to-image board — debuted across Meta AI, Instagram Stories, and WhatsApp. Meta is also teasing Muse Video ahead of a wider release, and dropped Muse Spark 1.1, a low-cost agentic coding model now on its new Meta Model API.
Chinese labs ramped up the competition. Tencent shipped Hy3, a smaller open-source model that punches above its weight, while Meituan fully open-sourced LongCat-2.0 — billed as the first trillion-parameter coding model trained without Nvidia GPUs. The trend: Chinese models continue to rival US labs’ model performance at a cheaper cost.
THE AI ACADEMY
How to turn your PDFs into a Claude skill

Go to Claude Code and upload a PDF you want to turn into a Claude Skill
Extract the main system from the PDF: the steps in order, repeated rules, mistakes to avoid, key questions, and what a successful result looks like
Open Claude and ask it to create a ‘SKILL.md’ using this prompt:
“Use the skill-creator to build a skill from the uploaded PDF. Activate when I ask Claude to [exact task]. Core steps: [paste]. Rules: [paste]. Mistakes to avoid: [paste]. Questions used: [paste]. Success looks like: [paste]. Do not use for: [2–3 unrelated tasks]. Ask me anything missing before generating the SKILL.md, then score the skill out of 10.”
Test the skill with different prompts to ensure it activates correctly and asks for clarification when needed.
Upload it in Claude through Settings → Capabilities → Skills → Upload
Now use your slash command whenever you want Claude to apply the PDF method.
PRESENTED BY CAESTRO
AI adoption is failing at your company because your team is stuck with tools that never learn. Caestro lets you hire agents like a regular employee and run them on company memory we build and maintain for you.
Hire with a real job description. Nothing to configure.
Works in Slack or email — a manager you name approves every output.
Caestro shadows, asks questions, and learns like a real employee.
Built by operators from Snowflake's first 50, Goldman Sachs, and Microsoft.
IN THE KNOW
What’s trending on socials today

Meme of the day
👀 Poster Backlash: AI-generated flyers are getting backlash online, with one viral post (3M views) saying they’re “polluting” real life. This article breaks down the graphic style that you may want to avoid.
👑 Meta Dethroned: How competitive is the AI space right now? Meta debuted its first image model ever two days ago — and has already been outranked by a new model from a research lab.
☎ Phone an AI: Video director Karen Cheng connected a vintage 1920s rotary phone to an AI agent, which then responds via a mechanical message board. The project is a subtle ad for Cursor, but is still getting praised for its creativity.
🚗 Model Shifter: An AI enthusiast built a digital gear shifter to manually switch through AI models. The comments agree it’s simultaneously useless and amazing.
🤔 Claude Reflection: Claude now has a new reflect feature that spins up a monthly recap of your usage statistics, with options to set quiet hours or break nudges.
PRODUCTIVITY
5 New & Trending AI Tools
📽 Runway: Generate realistic short videos with AI.
🤖 RunInfra: Optimize any open model for production with AI.
📁 Tabstack: Extract data and automate the web in one API call.
🔠 MixTranslate: Get translations from multiple AI models, side by side.
🤝 Maxworker: Turn your team’s context into next steps, reminders, escalations, and morning briefings directly in Slack.
* indicates a promoted tool, if any
PROMPT STATION
Learn any Topic
ChatGPT Prompt: Act as an elite teacher, curriculum designer, and learning scientist. I want to master [TOPIC] from [CURRENT LEVEL: beginner/intermediate/advanced] to [TARGET LEVEL] in [TIMEFRAME]. Create a complete learning roadmap with foundations, intermediate concepts, advanced topics, practical projects, exercises, quizzes, milestones, common mistakes, prerequisite knowledge, recommended resources, real-world applications, revision cycles, and progress checkpoints. Structure everything into daily, weekly, and monthly actions. Also explain why each learning stage exists and estimate mastery difficulty on a scale of 1 to 10.Friday Fun

Left:A, Right: B
IN CASE YOU MISSED IT
Our most-clicked story from yesterday
Learn how to download Claude Fable’s abilities into an operating manual that you can feed to other models. Prompts and instruction included.
Whenever you’re ready to take the next step
Check out our Top 125 AI Tools
Get better at prompting with our Top 1,000 Prompts
Learn how to use AI at work and get certified with our free course
Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.
What did you think of today's email?
Until next time — Zain, Theodore, & the Superhuman AI team.





