How I Built One Screen to Watch My AI Agents

I have AI agents running most of my day. Writing, researching, fixing things — across my laptop and a little server that never sleeps.

For months, I couldn't see any of it.

Work happened in the dark. I never knew what ran, what broke, or what was quietly waiting on me. So I built one screen to watch it all — a single dashboard I open from my phone.

Here's the idea, and how you can build a tiny version of it yourself.

An AI agent is just a program that does a task on its own — reads your files, drafts a reply, runs a check.

The moment you have more than one, a new problem shows up: you lose the thread.

• Which agent is running right now?
• What did it finish while I was asleep?
• Did one of them get stuck and need an answer from me?

You end up opening five terminals and three chat windows to reconstruct what your own system did. That's not automation — that's a second job.

The whole idea is boring on purpose. One page. Live. Openable from your phone.

Mine shows five things at a glance:

1. The budget — how much of my AI usage I've spent today (two percent, most days).
2. The work — how many tasks finished. Yesterday it was ten. I did zero of them.
3. Every agent — what's running, what's idle, what's stuck.
4. The files — every project, in one place, without hunting.
5. The inbox — and this is the part that still gets me.

Here's the move that made the whole thing feel alive.

When an agent isn't sure about something, it doesn't guess. It stops and asks me — a real question, on my phone.

I tap an answer. It keeps working.

That single behavior turns "a bunch of scripts I hope are working" into "a team that knows when to check in." It's the difference between trusting the system and babysitting it.

You do not need a server, a fancy cockpit, or months of work. Start with the smallest thing that removes the blindness:

Step 1 — Pick the one number you keep checking.
Usage? Tasks done? Whatever you keep re-opening a tab to see. Just one.

Step 2 — Write it to a plain text or JSON file.
Have your agent append a line every time it finishes something: what it did, when, and whether it needs you.

Step 3 — Show that file on one page.
A single HTML page that reads the file and refreshes every 15 seconds. No framework. No database. One file.

Step 4 — Add the "ask me" line.
When an agent hits a decision it can't make, have it write a question to that same file — and send you a message (a Telegram bot is ten lines of code). Answer, and let it continue.

That's it. One number, one page, one question. You can grow it later. The point isn't the cockpit — it's that you can finally see what your AI is doing.

Everyone's rushing to build AI agents. Almost nobody can see what theirs are actually doing.

Give your AI one screen you can watch, and the whole thing stops being scary. You're not hoping anymore. You're looking.

Start with one number today.

*Want the exact tools and the starter idea? Everything's free — grab it below.*