Ahoy, AInauts,

Welcome to a new edition of your favorite newsletter.

What is happening in the AI world? The current debate in Europe: Mistral, the EU's flagship AI lab, now also offers the Chinese open-weights model GLM-5.2 on its own platform. Is Europe giving up on building its own frontier model? Maybe...

But as always, we are not just about news. We put the tools to work.

Here are the three topics that kept us busy this week:

Let's go.

P.S. In less than four weeks, Berlin will host the BIG BANG KI Festival with 11,000+ AI enthusiasts and 350+ speakers. This is a German-market event, and its website and ticket flow are in German. AInauts get 50% off the festival pass - details in the third story below.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

This issue is presented with support from BIG BANG KI Festival.

As we wrote in last Thursday's issue, the AI employee wave has our full attention. Lindy, Viktor, Notion Agent - and now Grok Bot from xAI.

Last week we promised to get into the beta and report back. We are in, and here is our test.

In a nutshell: Grok Bot is the easiest AI employee setup we have tried. It feels almost magical.

If this is your first encounter with the concept: you work with different bots that you configure for specific jobs. You can start with predefined roles such as Chief of Staff or create your own. You provide context, connect the tools the bot needs, and train it on the job. Grok Bot follows that model, but the setup is faster and more autonomous than Claude Cowork and similar tools.

If you have avoided these systems because the technical setup looked intimidating, Grok Bot lowers that barrier significantly.

Download the desktop app. At the moment, you still need one of the more expensive plans, such as SuperGrok Heavy or Cursor Ultra/Teams Premium.

After signing in, click the plus icon and create a bot. We gave ours only one thing: the name "Google Ads Pro." Nothing else.

The bot immediately asked what we wanted it to do and inferred useful tasks and context from the name alone.

We chose Google Ads deliberately. There is no ready-made connector, so we wanted to see how well the system guides someone who does not already know the technical path.

As with Claude Cowork and similar tools, you can add plugins and grant access to your tools. The standard options are present: Gmail, Calendar, Drive, and more.

What stood out: connecting the X API takes a single click.

If you want a bot to monitor live events, community sentiment, or trends on X, that is genuinely useful. No other system currently makes this connection quite as simple. It helps that X and Grok share the same parent company.

Because there is no native Google Ads connector, we expected the setup to become technical. Grok Bot started with the simplest route.

Route 1: the built-in browser. Grok Bot has its own browser. You briefly take over, sign in to Google yourself so the bot never sees your password, then hand control back. From that point, it can inspect the account through the browser.

It works, and it is extremely simple. In about two minutes, your bot can start helping with your ads. The downside: browser operation is slower and somewhat more error-prone.

Route 2: the direct API. This is normally the technical route, especially with Google. Grok Bot coached us through it cleanly and provided direct links whenever we had to complete a step.

We completed the connection in four minutes. Another strong detail: the bot handles secure key storage itself. You do not have to keep credentials in a chat or create local files.

The first ad-account checks were very promising. The system is simply enjoyable to use.

Grok Bot combines capabilities we know from OpenClaw- or Hermes-style agents with an unusually simple setup. We can see it becoming a meaningful competitor to ChatGPT Work, Claude Cowork, and the other tools in this category. We are already using it more often.

To try it yourself:

Tell us how it works for you.

P.S. There is also a capable mobile app for assigning tasks to your bots while you are away from your desk.

P.P.S. Other tools have improved their onboarding too. Claude now walks you through Cowork with /setup-cowork, and ChatGPT Work also does a good job. Grok Bot currently stands out because the entry is so smooth.

You're spending 40 hours a week writing code that AI could do in 10.

While you're grinding through pull requests, 200k+ engineers at OpenAI, Google & Meta are using AI to ship faster.

How?

The Code newsletter teaches them exactly which AI tools to use and how to use them.

Here's what you get:

Sign up and get access to the Ultimate Claude code guide to ship 5X faster.

Join 200K+ engineers

Last week we noted that Anthropic now adds invisible watermarks to Claude-generated text. EU AI Act, disclosure duties, the full package.

One week later, reality looks like this:

An MIT-licensed GitHub repository designed to remove those markings gained 10,000 stars within days and had passed 14,000 when we checked. It targets Claude watermarks, Google's SynthID text, OpenAI signals, and C2PA or EXIF metadata in images and PDFs.

The pattern is familiar: large companies introduce a control, and the open-source community tests how easily it can be bypassed. A classic arms race, as long as the work remains legal.

Claude's text watermark is an invisible statistical pattern woven into word choice. It does not visibly alter the text and survives copy and paste.

The tool reliably strips metadata and hidden Unicode characters. That is the easy part. But the actual text watermark lives in the phrasing. To remove it, the tool has a second model rewrite the entire text.

The repository is candid about the limitation: there is no guarantee that detectors will fail afterward, and rewriting can reduce quality.

If you are comfortable with Python, the command line, and Docker, you can experiment with it. For most people, it is not yet a practical everyday tool. Coding agents can help with a prompt such as:

You can also run the result through detectors such as Pangram.

Our own testing shows the same limitation: none of these detectors is error-free.

This marking-versus-removal contest will probably continue for years without a clear winner.

With our partner BIG BANG KI Festival.

Our final find connects directly to last week's closer: the internet is changing, and AI is becoming a gatekeeper for visibility.

This section gives you two things: a practical AI-search tactic you can use immediately, and a substantial German-market AI event with a 50% AInauten discount.

Language note: BIG BANG KI is a German-market event in Berlin on September 16-17. Its website, ticket checkout, and the following event visuals are in German.

The festival has built a public Grounding Page. It explicitly says the page supports entity resolution, disambiguation, and retrieval stability for AI search and answer systems.

In plain English: this page is built less for human browsing and more for AI agents. It gives ChatGPT and other systems a stable, dated source with the festival's essential facts.

You should steal the move. There is even an open Grounding Page standard behind it. Create one URL with your core facts: who you are, what you offer, where you operate, prices, dates, and verifiable details. It costs half a day and may be one of the most pragmatic AEO improvements you can make.

Because its organizers understand the shift: machines will increasingly decide what gets found, while people still decide what they trust.

The program reflects that tension. Eckart von Hirschhausen is giving a keynote about the deepfake scam industry using his likeness to sell dubious products. Other sessions cover authentic photography as a counterweight to AI fakes.

That brings us back to the previous story. If watermarks do not solve the authenticity problem, what does? The more synthetic noise fills the web, the more valuable genuinely human, in-person interaction becomes.

BIG BANG KI expects 11,000+ attendees, 350+ speakers from Frank Thelen to OpenAI's GTM team, and five parallel stages at STATION Berlin.

If you want to attend, AInauts can use code KOOP-AIN-249 to get the festival pass for EUR 249 instead of EUR 499. Remember: this event and its ticket flow are for the German market and are in German.

Have a great time if you go, and tell us what stood out.

That is it for today. As always, thank you for reading.

See you next week.

Reto & Fabian from AInauten