Ahoy, AInauts!

Welcome to a new issue of your favorite newsletter.

The AI world is becoming more agentic by the day, even for non-technical users. We keep writing about it for a reason.

Grok Bot, OpenClaw, Lindy, Viktor, Hermes, ChatGPT Work, Claude Cowork ...

Now Meta is stepping into the ring with a fascinating angle: personal superintelligence in your pocket.

We also have a useful skill for you today, and OpenAI has released new versions of what we consider the best image model. You can try them right away.

Here is what is inside:

Let's get into it.

AI made PMs faster. Multiplayer mode is still broken.

A PM can summarize research, draft a PRD, and mock up a prototype before lunch. The hard part starts when the team has to decide what actually gets built.

Jira Product Discovery gives product teams one place to capture insights, prioritize ideas with consistent frameworks, and build living roadmaps stakeholders can rally around.

And because it’s connected to Jira, the context behind every decision stays with the work—so developers and their agents know not just what to build, but why.

AI helps PMs move faster. Jira Product Discovery helps the whole team build with confidence.

Get Started for Free.

We spend a lot of time thinking about AI employees. Last week, we looked at how they are becoming more mobile and increasingly get their own computer.

That is what happened with Claude Cowork Cloud, ChatGPT Work, and most recently Grok Bot. We love its concept because it is so simple and intuitive.

Now Meta is following suit. And we think this could appeal to everyone who never wanted to build their own AI employee.

A quick caveat: We normally cover things you can use today. Muse is not available everywhere yet; it is launching in the United States first.

We still think it is interesting enough that you should know about it.

The agent is called Muse.

Mark Zuckerberg describes the goal as "Personal Superintelligence", and Muse is the first step: an agent deliberately designed not for work, but for your life.

Buy movie tickets. Plan a weekend trip. Book a tennis lesson. Fill out the school-trip form. Turn a saved Instagram recipe Reel into a shopping list.

It looks like a messenger. There is no empty prompt screen, just a chat with someone doing work for you. It feels similar to Grok Bot.

And because Meta owns WhatsApp, Muse runs there directly. That is genuinely convenient.

It keeps working after you close the app. The setup differs slightly from other agents: every user gets a dedicated computer in Meta's cloud.

At the top of the chat, you can see what the agent is doing and where it needs your approval.

It has tabs that other agents do not.

It can spend money. Muse uses Stripe Link with single-use card numbers, so it never sees your actual card details. It still asks before making a purchase.

As far as we could determine, Muse currently follows the standard AI pricing model:

But will it always stay that way? From here on, this is pure speculation on our part.

Tabs such as a personal news feed are almost built for advertising. Ads could fund the product and keep it largely free, right?

That is Meta's entire business model. And Meta has the best advertising machine in the world, which would gain even more data.

For now, the official position remains: No ads in Muse, and your conversations are not shared with Meta's ad systems.

We can believe that, especially at launch. But forever?

Meta's own security post says this separation currently relies on policy rather than technical enforcement.

Meta employees can access your agent's computer. A confidential-computing version where only you hold the key is supposed to arrive later this year.

To be clear, we are not condemning an ad-funded model. It can be a great deal for many users, and as marketers, we love Meta Ads.

For comparison, OpenAI has shown ads in free ChatGPT since February. We are testing those ourselves for AInauten and will share more on Monday.

Our expectation: Meta will first focus on reach and adoption.

The assistant that knows you best is the position every major platform is fighting for. Monetization comes later.

And when it does, few companies in the world know that game as well as Meta.

Always-on agents did not suddenly appear this week. But apart from Grok Bot, most are still not trivial to set up.

Muse is the opposite. No terminal, no API key, and no setup video from us. 😄

It looks remarkably simple.

It is also the first major agent aimed at your personal life rather than your job. That combination is what makes it interesting.

Meta builds strong consumer products, is extremely well positioned with Meta Glasses, and already reaches billions of people. With Muse, it may finally have a compelling AI product of its own.

That is our early view, at least. We have not yet been able to test it in detail because of the US-first launch.

Fortunately, we have someone on the right side of the Atlantic for releases like this. We will update you as soon as we have hands-on experience.

The next wave of millionaires will be people who figured out how to make AI work for them.

The window to get ahead is still open. But not for long.

Here are 200+ proven ways to make money with AI in 2026.

Sign up for Superhuman AI, the free daily newsletter read by 1M+ professionals, and get instant access to all 200+ ways to profit from AI this year.

Claim your free list

Time for a quick change of topic and something you can use today.

You know the situation: You are working in Claude Code or Cowork on a proposal, a contract, or a piece of writing.

Halfway through, you want to know what GPT thinks. Or Gemini. Or the new model everyone is suddenly talking about.

Until now, that meant: copy the context, open another tab, explain everything again, and paste the answer back.

More opinions meant more tabs, and none of them shared context.

Or imagine you are working in Claude when OpenAI releases a model that supposedly changes everything. That has been known to happen 😄...

You want to test it immediately, ideally without paying for another pile of subscriptions.

We built a small skill for exactly that. It is called /ask, and it does one thing: asks another model without making you leave your current chat.

Language note: The screenshots and downloadable setup guide in this section are currently in German, but the workflow and commands work the same way in English.

For example, you write:

Claude turns that into a short briefing: what the task is, what has already been decided, and the exact question. It attaches the file you named and sends everything to a GPT model.

The answer appears directly in your Claude chat.

And here is the useful part: Claude evaluates the result. It tells you where GPT is right, where it disagrees, and what both models may have missed.

One window. Multiple AI models. Claude remains the orchestrator.

Instead of gpt, you can use gemini, grok, deepseek, or cheap for DeepSeek Flash, which costs fractions of a cent. Any model available on OpenRouter can be added.

Behind the scenes, the skill uses OpenRouter, with one key for all models.

Create an OpenRouter account, add $5 in credit, create a key, and copy it.

Then download the ZIP and paste this sentence into Claude Code:

Claude installs the skill and tells you the key is missing. Ideally, do not paste the key into the chat. Say only: "The key is in the clipboard." Claude can retrieve it directly.

Done. The full instructions and skill download are available here:

Language note: The download page and detailed instructions are currently available in German.

As always, anything you send through /ask goes to another provider. The skill excludes secrets and client data from its briefing, but you still decide what is appropriate to share.

Your Claude assistant can help you make that decision.

The question "Which model is best?" gets a new answer every week, and that answer is outdated a week later. No model wins at everything.

What we have noticed: When you can compare models side by side in the same chat, you learn their strengths faster than any benchmark can teach you.

After three days, you know which one to ask for writing, which one to ask for numbers, and which one to ask: "What am I missing?"

Try it.

To finish, here is something you can try immediately.

In our view, the best current image model comes from OpenAI. It just received an update, so we put it through a few tests.

In ChatGPT, it is simply called Images 2.5, and it is already available to everyone.

The API offers two variants:

The promise: more natural light, richer texture, and edits that change only what you asked to change.

So we tested it. Same prompt, same quality setting, all three models through fal.ai.

Photo: A craftsperson with a tablet in a workshop, morning light.

Image 2 produces a clean stock photo. Flare and Sunburst look slightly more convincing: sawdust, wood grain, skin texture, and light that appears to come through an actual window.

Flare even chose an over-the-shoulder perspective on its own.

Edit: To test image editing, we changed only the text on the tablet in the first image. Everything else was supposed to remain identical.

Language note: The requested test text in the comparison image is German by design; it reads "Moin AInauten."

All three models spell "Moin AInauten" correctly on the screen, while leaving the person and workshop untouched. Sunburst is the most precise.

The original screen content remains visible underneath. Flare reconstructs the screen instead.

Speed: In our fal.ai tests, Flare and Sunburst were four to six times faster than Image 2, including queue time.

Cost per high-quality image: roughly five cents for either new model.

The new model is already available to everyone in the ChatGPT app.

There are two new things you can test today:

We already loved OpenAI's image model, and this update is excellent.

But it is not a game changer because the previous model was already very good. Flare is enough for everyday work, but becomes too imprecise when details really matter. For that reason, we are going all-in on Sunburst.

That is it for today. Thanks a million, as always!

See you next week.

Reto & Fabian from AInauten