I Asked 1,000 Readers What's Broken About AI. Their Answers Explained Everything I Built.
Five months, one thousand subscribers, and the problem you keep describing without knowing it.
TL;DR:
I asked my community their #1 problem with AI. Few answered, but the pattern was impossible to ignore: people use AI without clarity on what it actually changes about their work.
That problem is exactly what I’ve been trying to solve — in five months of writing, and in the course I built.
Today I’m celebrating this milestone with a concrete offer for anyone who wants to stop drifting and start navigating.
A few days ago I asked a question on Substack Notes. I wanted to know: what’s the number one problem with AI, in your experience?
A handful of answers came in. But the pattern was too clear to ignore. Two stopped me cold.
The first: “For small business owners the number one problem is not knowing where to start. AI feels like a vast ocean and they are standing on the shore wondering which direction to swim.”
The second, longer and more precise: “The #1 problem with AI adoption is usage without clarity. People may be using AI tools, but many still can’t answer the questions: What is this for? How does it change my work? What am I allowed to use it for? What risks am I responsible for? Where does my judgement still matter? Who supports me when I’m unsure? That’s why usage doesn’t become adoption.”
I read those and thought: this is exactly why this newsletter exists.
Five months ago I ran an experiment on myself
The question was simple: can I actually transmit what I know about AI? Or is this the kind of thing you only understand by living inside it?
I started with no precise expectations. There were hard moments: weeks when the subscriber count seemed frozen, articles I’d worked on for days that went nowhere, the feeling of saying something important into a void. I questioned whether it was worth continuing.
Then a comment would arrive. Someone writing that an article had changed how they work. Someone forwarding a piece to a colleague with the note “read this, it matters.” And I’d keep going.
Today we’re at 1,000.
I’m not saying this to brag. I’m saying it because those Notes replies confirm something precise: the problem is real, it’s shared, and there’s room for someone who offers clarity instead of hype.
This isn’t a technology problem. It’s an orientation problem.
The six questions in that reply aren’t asking “how does a transformer work” or “which model is best.” They’re asking: how does this fit into my professional life? How do I know when to trust AI and when not to? How do I build a process that doesn’t break every time there’s a new update?
That’s the blind spot most courses and articles don’t actually address. They teach you to use ChatGPT. They don’t teach you to become the person who knows when to use it, how to verify it, and how to think with AI rather than just depend on it.
The problem with AI isn’t that it’s too complicated. It’s that most people use it without a map. And without a map, every update is a reset.
From User to Orchestrator exists to give you that map. Not an instruction manual for a single tool, but a structured path for understanding how to reason with AI - what to ask it, how to evaluate it, how to integrate it into real work without starting over every six months.
Reality Check
Ten new AI courses launch every week. Most of them teach you to use a specific tool and call it a revolution. The problem is that the tool they teach today will be superseded in six months and probably already has been. Courses that teach tools have an expiration date baked into the price you pay. The only thing that doesn’t age is the way you think with AI. Not the tool name. The method.
1,000 readers. One milestone. One concrete offer.
To mark this milestone, I want to do something simple: explain clearly what you get if you decide to support this channel as a paid subscriber.
Annual plan, €5/month (billed €60/year):
Access to all paid content: deep dives, operational analysis, experiments
25% off From User to Orchestrator and every future course, permanently
Ability to comment and discuss on all paid posts
Founding Member plan, €150/year:
Everything in the Annual plan
40% off the course and future courses (the code arrives in your welcome email, it’s a reserved benefit, not shown on the plans page)
A 30-minute 1:1 call with me every quarter, on your AI questions
On that last point, one honest thing: this call exists right now, for Founding Members. I don’t know for how long. When the number of Founding Members reaches a level I can’t sustainably manage, I’ll adjust the benefit or the pricing. There’s no artificial countdown. But there’s also no guarantee it’ll still be there in six months.
For everyone else: thank you.
Even readers who stay free and never upgrade are part of this. Free articles stay free. The newsletter continues. The work doesn’t change.
1,000 subscribers isn’t a finish line. It’s confirmation that it was worth starting.
See you soonk. Matija
I think expectations for an LLM are simply way too high, and all the marketing hype doesn’t make it any easier—but in so many ways, they’re capable of so much, only to fail miserably at the simplest questions (from a human perspective)!
That’s what Demis Hasabis says, too—and keep in mind, we’re still in the early stages, and it’s a daily experiment!
-> And the fact that it’s improving so much in such a short time leads one to assume that it can now handle even these simplest tasks, only to be “disappointed” again—and that’s usually not the fault of the LLM/agent but rather the user’s unrealistic expectations :)
As always -> the mistake is usually sitting just 50 cm in front of the screen! Error50
Just approach it with more patience, and it’ll get easier :)