Hey AInauts,

Welcome to a fresh edition of your favorite newsletter.

This week it's once again almost all about a shift we've been tracking for a while: AI moving from tool to genuine assistant.

First we look at new hardware built for exactly that, and then at the question of where you actually stand right now — including our current favorite AI agent.

Here's what's inside:

Let's dive in!

This edition is brought to you by:

We've been circling the same theme for weeks: AI is stopping being a tool you operate. And becoming more like a coworker you manage.

This week, Microsoft Build brought the hardware side of that story.

Microsoft and NVIDIA have jointly unveiled a new class of Windows PCs.

The idea: instead of burning cloud credits on every serious task, your machine handles it locally. On the device, not in the cloud.

Local AI has been around for a while, but it's now getting seriously powerful. To give you a sense of the scale: up to 1 petaflop of AI performance, 128 GB of memory, and the ability to run models with 120 billion parameters locally. That's a lot of firepower for something sitting on your desk.

Microsoft's own flagship for this is the Surface Laptop Ultra. The devices are coming this fall — from Dell, HP, Lenovo, ASUS, and MSI as well.

Cloud AI (the way we all use it right now) works like a casino.

You drop a coin for every image, every video, every agent task. Sometimes you pay directly, sometimes your SaaS app picks up the tab in the background. But someone always pays. Casino, because it's addictive. 😄

Which is fine — we do get something in return, after all.

Local AI flips that around. The intelligence runs on a device you own.

The first shift has been underway for a while: you ask, the PC delivers — via cloud AI.

Now comes the second: you ask, the PC delivers — locally and "for free."

The cloud stays important for the really heavy thinking. But everyday work on your machine will increasingly run locally.

Anyone who cares about data privacy should be pretty happy about this.

Also worth noting was the appearance of Austrian Peter Steinberger, the creator of OpenClaw.

Microsoft and Steinberger are working flat out to make OpenClaw Agents safe for enterprise use.

We find this incredibly exciting. That security piece is exactly why so much is happening right now. More on that in the last topic below.

Sure, the devices don't arrive until fall. And this is just the opening shot in the hardware race. Microsoft and NVIDIA have set the bar — hopefully Apple will raise it at its developer conference WWDC this coming Monday.

But you don't have to wait. Local models run on your machine today. With tools like Ollama, you can have a model running locally in half an hour — no cloud required.

We've written a lot about this — just last week in fact — or just search for "ollama" in our archive.

We can't wait to run genuinely powerful AI locally. This is going to be a whole new era for the PC!

One last question for you: Where do you actually stand with AI? And where does your company stand?

We work with a lot of teams inside companies, and we keep noticing the same thing: most people think they're using AI — but very few have any real sense of what's actually possible.

This week we stumbled across a genuinely great overview from Every.to.

It's essentially a ladder of AI usage, from the very bottom to the very top. Eight rungs, from simple chatbot to orchestrator — someone who runs entire teams of agents.

Eight levels feels like a lot to us. We'd say it's more like 3:

Our experience shows: 98% are stuck at Level 1.

The third level — the agent that works without you — has honestly been a tinkerer's game until now. Unstable, built for techies, pretty risky.

We've written a lot about it, but mostly told people to hold off.

That's exactly what's changing right now, as we touched on at the start.

Local compute power combined with the new security layers is making always-on agents affordable and reliable enough for the first time.

We think the moment has arrived where non-techies can start dipping their toes in. So expect a lot more from us on this going forward.

When it comes to this topic, the world is mostly talking about OpenClaw.

But the tool we're having the most fun with right now is Hermes Agent. It comes from Nous Research and is open source.

What makes it special: Hermes learns as it goes. From tasks the agent has handled well, it builds itself small skills. It remembers across sessions what you're working on and who you are. They call it "the agent that grows with you."

Hermes is optimized for local LLMs running on your machine — your data stays with you, no tracking. And it lives where you already hang out: Telegram, Slack, WhatsApp, email.

Until now, Hermes was a terminal thing — not exactly accessible for regular folks. This week the native desktop app dropped, in a first preview for Mac, Windows, and Linux. Which means anyone can finally just try it out!

We're playing around with it a lot ourselves right now — and more topics including deep-dives are coming soon.

Ask yourself quickly: Which of the three levels are you on right now?

If you're stuck on level one, your next move is to jump into files. Let AI actually work alongside you instead of just answering questions.

Grab Claude Cowork, Codex (free), open a folder, and get going.

If you're already on level two, right now is a great moment to test an always-on assistant for the first time.

Just give Hermes a spin — no terminal fiddling required anymore.

And heads up: A higher level isn't automatically better, by the way.

Nobody needs to become an orchestrator at all costs. The skill is choosing the right level for the task at hand. But knowing the upper floor exists — and that it's opening up right now — doesn't hurt.

→ Take a look at Hermes Desktop

Alright, that was a bit more forward-looking than usual today. More things to try out coming soon!

See you this weekend.

Reto & Fabian from the AINAUTEN