“Last week a US company asked us to teach them how to code with Claude. We didn’t prepare a presentation. We opened the laptop and showed them how we work. It was the best session we’ve ever delivered.”
Last week, a US home-improvement company asked us to teach them how to program with Claude Code. They wanted a workshop, a structured training, something they could fit into their digital transformation plan with a nice corporate name.
We didn’t prepare any slides. We didn’t set up a polished demo. We opened the laptop and showed them how we work. As-is. With our real environment, our real projects, our real mistakes.
It was the best session we’ve ever delivered. And what surprised them the most wasn’t any advanced feature or technical trick. It was the simplicity.
The thesis is this: the best way to teach AI isn’t to explain it. It’s to use it in front of the people who want to learn.
The AI consulting problem nobody says out loud
The pattern repeats in every company we’ve seen hire AI consulting. The consultants arrive with impeccable slides. The team leaves motivated. Inspired. They go back to their desks on Monday morning.
And they don’t know where to start.
The real gap is this: between “understanding conceptually what an AI agent is” and “using it to solve the concrete problem I have on Tuesday at 10am” there’s an abyss. An abyss you don’t cross with slides. You cross it with practice.
And here’s the part nobody says out loud: most AI consultancies teach AI they don’t use. They sell theory because they don’t have practice. They build pretty demos because their real workflow doesn’t exist — or isn’t solid enough to show.
If you asked your consultancy to open their laptop and show you how they use AI in their own business, could they do it?
What we did: open the laptop and work in front of them
No slides. No prepared demo. No script. We opened our work environment — the same one we use every day — and started working in front of them. We showed them our programming protocol with Claude. Specifically:
- How we structure instructions:
CLAUDE.mdfiles where we define context, rules and workflows so the agent behaves consistently. It’s like giving an operations manual to a teammate. - How we iterate: you don’t expect the agent to get it right on the first try. You provide context, review the output, correct, and iterate again. It’s a workflow, not a magic button.
- How we organize context: meaningful folders, clean documents, distilled information. If your data is chaos, it doesn’t matter how good your model is.
- How we review and correct: the agent makes mistakes. Sometimes a lot. The difference is having a clear process to detect and fix them fast.
What surprised them most was the reaction we always expect but it never stops being surprising: “Is that it?” Yes. That’s it. No complex infrastructure. No proprietary platform with a six-figure license. Just a clear, repeatable and — above all — simple workflow. Before the session ended, they were already replicating it with their own projects.
Why it works: the “I can do that too” effect
When you watch someone work with AI in their real day-to-day — with the agent getting it wrong, with iterations, with moments of “this isn’t what I asked, let’s rephrase” — it demystifies the process.
In a prepared demo everything goes perfectly, but you learn nothing. In a real workflow you see the full process: attempt, error, correction, result. That’s what makes it replicable.
If your AI workflow needs a 50-page manual, it’s not a protocol — it’s bureaucracy. The best workflows are the ones you can explain in 10 minutes and someone can start using that same afternoon.
The uncomfortable question
If the consultancy teaching you AI doesn’t use AI internally — to manage projects, to write code, to organize knowledge — what exactly are they teaching you?
The best AI training isn’t a two-day workshop with a certificate. It’s watching someone who uses it for real, working for real, solving real problems. Mistakes included.
If you want to see how we work, you only have to ask. No slides.
Related article: Using the latest AI model is not a competitive advantage — why data matters more than the model.