Thirty people and their first agent

May 14, 2026

Last month we ran a small hands-on workshop for people in Armenia who wanted to build their first autonomous agent.

The idea was simple: bring your laptop, buy a cheap VPS, and in 3 hours try to get your first agent running on your own server.

Just setup, errors, debugging, and a lot of “wait, why doesn’t this work?”

Three days before the event we had around 50 sign-ups, which was more than I expected. I knew not everyone would come, but still, I couldn’t support that many people alone.

So I asked in the Telegram group if anyone could help as a mentor.

Five people said yes.

Nobody got paid. They just came to help.

Before the workshop, people had homework:

- buy a VPS, we used Hetzner, around €5
- get Claude or GPT access
- come with a laptop ready to work

On Saturday around 30 people showed up.

We split into 5 groups, one mentor per group, and spent the next 3 hours trying to get everyone unstuck.

Some people were fighting with SSH.
Some were setting up their first server (it took 1 hour for non tech folks to buy vps droplet).
Some were already asking how to connect the agent to their actual work.

That was probably the best part: people weren’t asking abstract questions about “the future of agents.”
They were trying to make something run, suffering together from bugs, tech problems and experiencing pain from working with CLI for first time.

We also covered the practical side of multi-agent systems: what runs where, what breaks, how agents talk to each other, and why “just connect a few tools” is usually not enough.

At Empy.ai, we’ve spent the last year building realtime voice agents, and a lot of our useful lessons came from annoying bugs, not clean blog posts.

So the goal was not to teach “AI agents” as a big concept.

The goal was to give people enough practical context to stop being spectators and start building.

By the end, every group had something running on their own VPS.

Not perfect, not even production ready.
But real enough that they could continue after the workshop.

The whole thing was pro bono, to meet non tech people and give them glimpse of AI native reality most of tech founders live in.

I just wanted more people in Armenia outside the usual dev bubble, founders, designers, scientists, and real-sector entrepreneurs, to get their hands dirty with agents instead of only reading about them.

I left pretty convinced that the best AI workshops are basically group debugging sessions with a little bit of theory in between.

People don’t need another talk about what agents might do someday.

They need someone next to them when SSH breaks, cry together when they can't pass KYC at VPS provider.

At the end of the day, misery loves company :D