Sponsored by
Ahoy, AInauts,
Welcome to a new issue of your favorite newsletter.
OpenAI tested how well a new model could hack. It promptly found a way out of the test environment and started exploring beyond it. Unsettling, but these are the kinds of stories we will have to get used to.
Today, however, we are focusing on what AI can do for us. We have practical workflows, a useful creative platform, and one less exciting legal topic 😉.
Here is what we have for you:
🎥 Record a skill: Show it once, and your AI can do it forever
🛰️ Magnific: One login instead of five tools, and why AI video does not have to be expensive
⚖️ Starting August 2: Do you have to label AI-generated content?
Let's get started.
Save 10+ Hours a Week With 37 Claude Prompts
Every manager faces the same situations before lunch: a message to land, a meeting to run, a hiring call, a report due. The AI Report built 37 Claude prompts for exactly those moments, organised by the situations every manager faces.
Copy the prompt, fill the brackets, run it in Claude, and get back 10+ hours a week. Oh, and it's free.
All you have to do is subscribe to The AI Report, a 5-minute daily AI brief read by 400,000+ business leaders at IBM, AWS and Microsoft, and the full prompt pack lands in your welcome email. The newsletter and the prompts, both free. Subscribe and grab both
🎥 Your AI Employee Watches and Learns
We are seeing more and more people in our community, thankfully, working with Claude Cowork or ChatGPT Work, also known as Codex.
That is a big step: moving out of the chat window and toward a real AI employee.
Yet too few people use skills with their AI agents.
Skills are a genuine game changer.
A skill is simply a task that you teach your AI assistant once so it can handle the job on its own afterward.
The report you build every Monday. The invoice you always file in the same place. The proposal that follows the same structure every time.
Why do so few people create them?
Our theory: people do not take the time to think through a task and describe it in enough detail for an assistant to turn it into a reliable skill.
A useful skill takes work. Most people never get around to doing it.
Now there is a much easier route, so the excuses are running out.
You Do the Task Once, and Your AI Watches
You can now share your screen with an AI so it can watch and listen while you work.
OpenAI moved first in June. The Codex app includes Record & Replay. Start recording, complete a recurring task on your computer, such as filing an expense report, and stop.
Codex turns that demonstration into a skill it can run again with new data. The feature is not available to us yet.
Anthropic followed this week. In Claude Cowork, the feature is called Record a skill. You record your screen and explain what you are doing out loud.
Your clicks, inputs, and explanation all become part of the skill. You teach the AI what you are doing, why you are doing it, and where the exceptions are.
Speaking out loud is the real trick. The tool sees more than your clicks; it understands the intent behind them.
Visual note: The following Claude Cowork screenshots use German interface labels.
Claude Cowork's Skills menu with German interface labels.
How to Try It
Choose a task you genuinely perform often and that follows the same steps each time. Do not start with the most complicated task. Start with the most boring one.
In Cowork: Customize → Skills → Add → Record screen. The interface labels may vary by language.
Complete the task normally while the recording runs. Explain each step as if you were onboarding a new colleague.
The Record a skill dialog in Claude Cowork. The screenshot uses German interface labels.
Next time, you run /yourskillname, and the workflow starts automatically.
Our Take: Let AI Do the Work
There are fewer excuses now. Make your AI employee genuinely useful by showing it the repetitive work you want off your plate.
Demonstrate the task once, explain the context, and delegate it for the future.
The feature still has limits. You will not delegate everything, but it is a strong fit for the boring, repeatable work.
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.
🛰️ One Login Instead of Five Tools, and Why AI Video Does Not Have to Be Expensive
Another trip down our AI video rabbit hole. Today we are revisiting a platform that is useful for more than its AI features: Magnific.
Freepik is now called Magnific.
The new platform combines the former stock library, the Magnific upscaler, and current AI models under one roof.
Magnific also includes many other image, video, audio, and design tools.
Why Does This Matter?
AI video is expensive and often slow. You pay per clip and per attempt.
It may take three or four tries to get one usable result. Suddenly, you have spent $20 or $30.
But you do not always need to generate everything from scratch. Plenty of scenes already exist as real stock footage.
Stock costs nothing extra and is available immediately.
We give our AI employee one simple rule: Use stock for everything that already exists. Generate only what does not exist, such as your product or a very specific scene.
Magnific is one of the few platforms that combines both in one place. Better still, you do not even need to sign in yourself. Your AI agent can operate it for you.
Magnific connected as an MCP tool.
We connected our agent to Magnific through MCP and had it complete four tasks without opening the interface ourselves:
Create the same astronaut once with stock footage and once with AI. Both looked good. Stock was instant and cost nothing; the AI version took five minutes and used credits.
Upscale a pixelated logo in Magnific until it was ready for print again.
Convert the same logo into an infinitely scalable SVG vector graphic.
Edit stock footage with Google's Omni model and change it to our brand color.
The stock original is on the left; the Omni edit in our brand color is on the right. One prompt, done.
Ten minutes, roughly 5,000 credits from our allowance, and every asset was saved in one folder.
If you prefer clicking through the tools yourself, Magnific has a strong interface. It is almost a Canva alternative, with templates, vectors, and ready-made workflows.
Magnific's tools for images, video, audio, design, 3D, and stock media.
We find Magnific interesting for two reasons.
First, newer tools without years of technical baggage are often built AI-first. That makes them easier for AI employees to operate.
Second, Magnific means we do not have to generate every asset from scratch. Stock images, videos, vectors, and templates are ready immediately.
That saves tokens and often saves time.
Magnific's stock library includes vectors, photos, videos, and templates.
⚖️ Starting August 2: Do You Have to Label AI-Generated Content?
We will finish with a quick look at a deadline that is getting closer: the transparency obligations in the EU AI Act, especially Article 50.
As you may already know, we are not lawyers, and this is not legal advice. It is not our favorite topic either.
European Commission information about labeling AI-generated content. The screenshot uses German interface text.
We are still working through the details ourselves and discussing them with lawyers and other specialists.
Treat this section as a reminder that a deadline is approaching and you need to check your own use cases, not as a final legal opinion.
Based on our current reading:
Text is surprisingly straightforward. The disclosure obligation mainly concerns AI-generated or manipulated text published to inform the public about matters of public interest. Even then, it may not apply when a human has reviewed the content, exercises editorial control, and takes responsibility for it. That should exclude many everyday AInauten use cases.
Images, video, and audio are the stricter category. If you publish realistic synthetic or manipulated content that qualifies as a deepfake, you generally need to disclose that it was artificially generated or manipulated. Artistic, creative, satirical, and fictional works receive more flexible treatment. There is no simple “a human reviewed it” exception for this category. We are still determining how far the rule reaches in practice.
Interaction: If your website uses a chatbot, people need to be able to recognize that they are interacting with AI. Most tools already make this clear.
Overall, we do not think the requirement has to be dramatic. A small label should not hurt much once it becomes common and people start treating it like the familiar sponsored-content notices on YouTube.
The boundary is still interesting. A person can heavily alter a body in Photoshop without automatically triggering an AI disclosure. If AI creates or manipulates realistic content, a disclosure may be required.
Language note: The following additional articles are in German.
That is it for today.
As always, thank you for reading, and see you in the next issue.
Reto & Fabian from AInauten
