Does an Agent System Need an Interface?

My own agent system started life on a rented server. No screen, no buttons, no window. It ran in the background and did what I handed it.

Why I started this way

I assumed a good automation doesn't need to be seen. No interface means nothing to maintain and nothing to design, just the process. It felt simple and fast, and it was, as long as it had little to do.

What happened

Then it took on more and more. It researched topics, gathered and analysed information, wrote drafts, prepared things. Everything it produced went through me before it went anywhere. That principle hasn't changed: I automate the input and the output, never the judgment.

But to do that, I had to see what it had done. Approve it, send it back, rewrite it. And you can't look inside a system that only runs in the background. So I built it a PWA, an app that runs in the browser.

Now a native iOS version is being built. That way Siri can connect to it, and I can reach things the browser doesn't allow.

What surprised me

That the interface was never about convenience. It was about control. The more the system does, the more it matters that I can see what it did. A system I can't see into will sooner or later start deciding for me.

What I'm taking from this

An interface isn't decoration on an automated system. It's how the system fits into my day, and how the decision stays with me.

The experiment is still running. I'm watching whether the native app and Siri change how often, and how, I look in. If you're automating something yourself, plan on day one where you'll see and approve what it does.