skip to content
Petr Reichl
amazon

I met four young guys at a café in Bali. They sat at one table, each with his own laptop, VSCode open, building something. The energy caught my attention. It was obvious they were building something together. That great startup energy when an idea is taking shape. I had a nice view from the next table, so I watched them for a while. They were running the classic developer flow: documentation on one half of the screen, VSCode on the other, everything written by hand. No AI. No Claude Code, nothing like that.

I started talking to them to make sure that was really the case and to learn more about what they were doing and how. Young guys, students, cheap laptops, a lot of enthusiasm. And really no AI. No Copilot, no ChatGPT.

I thought about it a lot afterwards and looked up wage numbers in Indonesia. I pay a few hundred dollars a month for AI that gives me, I’d dare say, far more productivity and capability than all of them put together. The base $200 for a Claude plan is, in Western prices, somewhere between lunch and cheap hosting. A pretty meaningless line item. A junior programmer in Indonesia outside Jakarta can earn 5 to 9 million rupiah, so roughly $300 to $560. And there it stops being a negligible amount. It’s almost a third, or even two thirds, of what such a programmer takes home. And keep in mind these subscriptions are heavily subsidized by the LLM vendors, and sooner or later that’s going to have to change.

This is where the uncomfortable part starts. For young guys somewhere in Indonesia, access to interesting technology and projects is much more limited than it is for a similar programmer in Europe or the US. Learning from senior people, using cloud technology, that kind of thing — it already creates a real gap. Now AI and coding agents have been added on top. While I spend my time turning my company over to agents, with the goal of having agents run as much of it as possible, they spend their time writing CSS and building components in react.js.

A serious gap is forming here, and it’s going to be very hard to cross. Maybe in the end these guys will be cheaper than the AI agents, which will get used for the more important work. I spend my days teaching agents to do work for people. They spent their days doing that work by hand — exactly the work agents will swallow first. We’re on opposite ends of the same wave. The difference is that I get to pick a side. For them, it’s probably not that simple.