I’ve recently come across several comments and posts along the lines of “AI/vibecoding workshops are snake oil” (for example here). Allow me a few words, because naturally it’s on my mind whether I’m selling snake oil myself.
I’ve been doing agentic development (or vibecoding, if you prefer) for about 2 years, and I worked with Claude Code back when it was still a non-public version. So on one hand I don’t have much experience, but on the other hand I’ve been doing vibecoding for as long as the technology has existed.
I’ve been in software development all my life, but the last time I personally wrote a large, functional piece of code was 30 years ago, somewhere on the border between Fortran and Pascal. Since then I’ve only led development teams and have practically no time for coding. Mostly I deal with things like “who didn’t show up for work,” and when I do deal with an architectural decision, it’s basically in the style of “you two responsible for this, bring me proposals and defend them, you’re the smart ones here.” That’s the first point.
That’s enough of a recap of my life experience.
I thought for a long time about whether to get into training people in agentic development. First, I was a deputy mayor of my city at the time and took it as an occasional change of pace. Second, I quickly found out that I know quite a bit more than others. And what convinced me was seeing how much it helps companies.
What “training AI” means
Second, what does it mean to train people in AI? I always ask customers whether we’re going to inspire people or teach them. You can’t do both at once. When we come to a development company where people are standoffish toward AI, our role is to show them where it will help and to explain to management what it won’t do. The first is inspiring, the second is teaching, and you can’t do them at the same time. When you’re going to inspire someone, they don’t need to be flooded with details, because they’d drown in them; they need to see how it works. When you’re teaching someone, they already know it works, but they need the details.
Third, even agentic development is a very broad topic today. And I have to admit that I practically don’t tune Claude Code’s configuration parameters. I don’t adjust how many subagents Claude Code launches, or other environment variables. I leave them alone, because I’ve found that what Anthropic ships is the best setup for 90% of cases. We can google the rest, because I at least have a general idea that you can set the number of retries for the watchdog when you need to. That isn’t essential for development. If there’s something I focus on, it’s introducing agentic development up to an intermediate level, which today means about 8 hours of workshops and about 2 months of independent, active practice of what you’ve learned, i.e. applying the knowledge you’ve gained in real work. Workshops have a hard time moving you beyond the intermediate level, because that means specialization, which few people here understand, and few in the world in general. So you learn by trial and error. I can hardly advise you on how to run Claude Code in Bedrock…
How to choose the right workshop or training
In general, if you’re choosing a workshop or course, I’d suggest proceeding like this:
- Do you need to be Inspired or Taught? These are two different disciplines; a good speaker with nice examples is better than a technician pouring out unfamiliar acronyms.
- Have you seen the instructor speak somewhere, and did it make sense to you? If not, they probably wouldn’t suit you.
- If you want to learn, go for specialization. Don’t take a course for developers that introduces you to Claude Code, Codex, Cursor, Copilot, RAG, RL and who knows what else in 8 hours. You’ll learn a little about each and it won’t be of any use to you. When your project needs RAG, RAK or RAKE, the AI agent will tell you and explain it on request.
- When you go for specialization, ask for working procedures, workflows. How do you maintain documentation or CI with AI?
- When you’re already going for the next workflow, order custom training, or don’t be disappointed that a general workshop doesn’t answer questions specific to your company.
- An online workshop is always weaker than an in-person one: the instructor lacks contact with the audience, and you lack discipline.
References are a chapter of their own. I envy websites that announce “hundreds of trained companies,” and at the same time I wonder what that means. For example, ten thousand people have completed our online AI school; is that the same kind of number? Conversely, an individual intensive training was done for a few dozen companies, so did a person I’ve never heard of manage more? Hundreds of people came to my personal trainings, and about a thousand different people came to Vibecoding Talks; which of those is the number of people trained? So take references very loosely. The fact that someone from the Office of the President came to Vibecoding Talks doesn’t mean we train the Office of the President. Accept a reference only if you can go there and ask what it was like and how it went. By the way, we don’t put references on our website, because I respect our clients’ privacy.
And the money?
A good half-day workshop is two days of work for the instructor, so calculate how much you value your own two days. That’s where the prices come from. For a company, count on tens of thousands of CZK, so that it’s worth something. About 4,000 CZK per person is sustainable for a fuller workshop; 4 participants are more of a loss for a good instructor. That complicates the situation: these days anyone can sell training, and for someone who wouldn’t earn more at McDonald’s, 4 people is a profit, but you won’t get quality. That’s a competitive point, but also an incentive for you to think about whom you go to for a workshop.
The good thing is that there’s plenty to choose from. Zandl’s golden rule? A good instructor should give you a kick. A good training or workshop should shoot you up to the intermediate level. Between pros and seniors, you have to grow on your own, through careful collecting, sifting, and applying your experience.
And one thing is sad. Workshops are hard to compare, because few people attend several of the same kind. Several times people came to my workshop Claude Code for Developers and told me they took it as a more advanced workshop, because I don’t go through the absolute basics. For that reason I also started treating my second workshop, Selfimproving Loops, self-improving loops in Claude Code, as a more advanced follow-up, even though the first hour overlaps a bit. In general I’m curious what others are saying out there, and I monitor it a lot on YouTube, where Hypata gives me a daily summary of what’s being discussed in YouTube videos. Those are foreign experiences, though; in the Czech Republic I’ve only attended Petr Ludwig’s session, which was very nice, but very different from what I do. But don’t ask me how I compare with other instructors. I don’t know; I haven’t seen them.
To sum up: this industry is the same as any other. You’ll find swindlers, people who take it seriously, and enthusiasts who don’t realize what they don’t know. Over time, personal brands and brands you can lean on will crystallize, and in the words of a classic, “let fall away what is hesitant, let fall away what is shaky.”