An agent that watches the web on its own — and texts you first when the thing you're waiting for shows up

Here's the idea in one line: you hand an agent one task — "watch for tickets to this show" — and forget about it. And it, on its own, while you sleep, goes online every hour, decides for itself whether something genuinely new appeared, and texts you first: "they're up, here's the link." You don't poke the bot — the bot reaches out when there's a reason.

And here's what's new. Almost every bot is reactive: it waits for you to write. Doing the opposite — having something watch on its own and nudge you — used to mean writing a parser by hand, putting it on a schedule, storing the previous state, diffing it, and sending a notification. A pile of code for a single line: "it's here!" At Google I/O in May 2026, Google showed agents that run in the background 24/7 and reach out on their own when something changes — background agents in Search and the personal Gemini Spark agent that keeps watch and pings you only with what matters. The idea of a personal background watcher went mainstream — and you'll build your small version over a weekend, because now the agent itself does both the looking and the deciding whether to bother you.

Why this one

Everyone has their own "tell me the second it shows up." A console restocks. A ticket price drops. A flat opens in the right neighborhood. Tour dates get announced. By hand that means opening a tab ten times a day and missing it anyway. A watcher that keeps the post on its own and texts first takes it off you entirely. You'll actually use this yourself.

And there's less "magic" here than it looks. The tedious part — "watch on its own and judge whether it matters" — is now the agent's job. Your app just gives it a schedule and a channel to write to: a Telegram bot or a push.

What you'll learn

• Reactive bot vs proactive agent. The first waits for a command. The second reaches out on its own. You'll feel that difference firsthand — and get why the second one feels "alive."
• Why an agent needs search. To "see whether it changed," the agent goes to the web itself — that's tool use. You'll hand it those arms and a schedule to use them on.
• How not to spam yourself. The whole focus: the agent compares against last time and writes only when it's really new. You'll learn to store "what was there before" and give a sharp rule for "this one counts."

A ready starter prompt

Don't ask for "watch everything and ping me if anything happens" — it'll bury you in noise. Give it exactly one goal, a rule for "new," and a ban on false alarms:

A strong prompt leaves no room for noise: one goal, a clear rule for "new" (status flipped), a precise channel, and an explicit ban on empty messages. The watcher stays quiet until there's a real reason.

What you end up with

You've spent a week waiting on a console restock and you're tired of refreshing the tab. You set up the watcher and close the laptop. Thursday at 3 a.m. your phone buzzes once: "in stock, $599, here's the link." You tap through — and it really is there. You didn't open a single tab all week. An agent kept watch for you and reached out exactly once — when it was worth it.

Start with one run and one goal — and you'll have a thing that keeps watch in your place and finds you on its own when the thing you wanted shows up.

Source: Google I/O 2026: agents that run in the background and reach out on their own (Google Blog)