🧙🏼 How I run multiple AI workflows at the same time

SUBSCRIBE | ADVERTISE

Howdy wizards,

Welcome to the 489 new subscribers who joined this week.

Today’s specials:

• OpenAI is entering Europe, building it’s GPU gigafactory in Northern Norway

• Behind-the-scenes of my everyday workflow with AI

• Winners from Bolt’s recent hackathon

• ..and more

Here’s what’s brewing in AI.

DARIO’S PICKS

OpenAI is building a 100,000-GPU fortress in Norway

OpenAI is launching a massive data center (Gigafactory) with 100,000 NVIDIA GPUs in Northern Norway, expected to be ready by the end of 2026.

Stargate Norway is the first partnership for OpenAI in Europe under the 'OpenAI for countries' program. It will be a joint venture between Nscale (AI infra provider) and Aker (a Norwegian energy company), and run on renewable power. Reportedly, local businesses and researchers in Norway will get priority access to compute, while excess heat generated from the facility will go to power nearby businesses.

‎ Why it matters‎ ‎ OpenAI has entered the European chat.

Norway brings cheap renewable energy + a cold climate; ie great conditions for running and cooling down power-hungry GPUs. And could benefit the local economy with new jobs, innovation, etc.

Who, in practice, will actually control & benefit from the infrastructure built under these joint ventures? Norway will be an interesting case study to follow.

IN PARTNERSHIP WITH SHUTTERSTOCK

Download our guide on AI-ready training data.

AI teams need more than big data—they need the right data. This guide breaks down what makes training datasets high-performing: real-world behavior signals, semantic scoring, clustering methods, and licensed assets. Learn to avoid scraped content, balance quality and diversity, and evaluate outputs using human-centric signals for scalable deployment.

Download the guide now

UP CLOSE

This section is about how I’m using AI from week to week, as well as practical tips & tricks I discover and actually use.

How AI helps me juggle tasks while I stay in flow

I spent most of this week cleaning up the codebase for the analytics app I’m building in Cursor (yes, code is written by AI and AI cleans it up). I’m probably going to make a deep dive on refactoring AI code soon, as it’s a big unlock if you want to keep adding complexity and still have a sense of how your app works.

Anyways—last week I told you about how I’m vibe coding as a background activity, and I thought I’d give you a behind-the-scenes peek at how I multitask without really multitasking.

Hear me out.

I almost throw up when I hear the word multitasking. Frankly, I’m the kind of person who needs extreme focus on one thing to be able to get anything done.

But with AI, I can effectively run a bunch of processes at the same time without losing focus. Things I spend a while configuring, but once they’re up and running I mostly just check back on them once in a while. At any given time, my computer will be loaded with a bunch of processes like this.

Here’s a snapshot from Tuesday and the stuff I had running simultaneously on my Mac (illustrated above):

• A full coding workflow (all the green arrows) with Cursor running Claude Code in the terminal. This requires my full attention at specific times – whenever it’s finished doing what I asked it to. Reddit also taught me an easy way to get notified when Claude Code is done, which is helpful when you’re working on other things on the side.

• ChatGPT Desktop running with a single purpose – being a dictation box. For some mysterious reason Cursor doesn’t have this natively, so I use the ChatGPT instance for simply transcribing my instructions and copy/pasting them into Cursor. Make no mistake: this is likely the most underrated feature to your workflow, regardless how you use ChatGPT. There’s dedicated dictation tools out there, but I don’t want yet another subscription—so I consider this efficient enough for now.

• A good ol’ Apple Notes with upcoming tasks I’m going to give the AI; all voice recorded with ChatGPT (like shown above). The reason I don’t simply pour all the instructions to AI in one go is that I’m building a complex product and I want to test things at each meaningful progress step. Letting the AI have a go at a ton of different instructions at once when vibe coding makes a mess.

• GitHub Desktop. Vibe coding up anything more complex than a tic-tac-toe game necessitates the use of GitHub or some other kind of version control. Some vibe coding platforms have built-in functionality for this, but the more advanced ones don’t. In any case, you need a way to restore progress when the AI eventually messes up; it will mess up all the time, so there’s zero getting around this step. If you’re new to it, spend an hour of your life watching a beginner’s tutorial on GitHub on YouTube—you’ll gain that time back plentifold.

• Custom GPT for lead gen. I’m using a custom GPT I made in ChatGPT to make it easier to reach out to relevant sponsors for this newsletter. The GPT lets me drop a list of company names or screenshots of their logos. Then, without further instructions, the o3 model does research on each company to find their website, as well as the name + LinkedIn URL of a decision maker in the marketing department. Then outputs it all nicely in TSV format which I paste straight into a Google Sheet. It spends 20-30 seconds per company so if I paste 30 companies it’s done in ≈15 mins. I gathered info on 1,000 companies on Tuesday.

• Generating design assets in ChatGPT. I’m a big fan of 4o’s image generation. And when I’m not using it to turn myself into various cartoon characters, I’m actually creating image assets to use in my business. For example all the icons and images illustrating this very newsletter. I really enjoy the ability to use a reference image, that’s how I get the character consistency of my wizard mascot. I also like to find cool and unique scenes that I can place him into and aesthetic photographs to use as inspiration. Since the results are kinda random I typically cue up 5-10 images and just let Chat do it’s thing, then select or iterate on my favourite versions.

Phew! I don’t know about you, but that’s enough stuff running at once to make me a bit dizzy. I’m not going to pretend that I work like this all the time, every day.

A workflow like this, while fairly easy to manage, still has an element of task switching to it which isn’t how I want to work all the time. Also, it doesn’t pair very well if I have a case of brain fog – or worse – am undercaffeinated. So I try to work like this only when I’m feeling highly alert and up for the challenge—I get a lot done during those times.

Then again, productivity isn’t just about how much you get done – especially in this age of AI – but more about what you get done. I think it’s more important than ever to disconnect frequently and be distraction free for longer periods of time, to let fresh perspectives arise.

In other vibe coding news:

• Winners from Bolt’s vibe coding hackathon. My favourites are Bored? Opposite! and ModelMash. I’ve added links to the top 10 winning apps in my guide to the top vibe coding tools (under Bolt):

• What type of apps should be vibe coded by non-engineers? Simple sites, prototypes, and internal tools are easy targets.

• Google’s head of Product says we’re moving from a writing-first to building-first culture. With AI, engineering resources are less scarce and development goes faster; non-technical people can finally show their ideas instead of talking about it, and a lot less get lost in translation.

• Anthropic is working on finding the right price point for their insanely popular command line coding tool Claude Code. Some people have been using it to run coding agents 24/7, but now the free-for-all is over and they’re introducing weekly rate limits. Reportedly, that affects less than 5% of users. Very likely, that’s me and probably a bunch of you too.

• To help with the upcoming rate limits, I found a neat Claude usage tracking app which lets you track Claude API or paid subscription usage; it can distribute workloads if you have multiple accounts.

PS last week I asked how many of you have tried vibe coding tools at this point (Cursor, Bolt, Lovable, Replit, etc.). Over half of you have tried it or are actively building with it! That’s awesome. For those of you in regulated industries who told me you can't use it at work: start a hobby project—build a personal productivity tool or that game you wish existed. These skills can open doors, and when opportunities arise, you’ll be ready.

THAT’S ALL FOR THIS WEEK

Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.