My AI agent lives in a data centre. It should live next to my router.

Say you ask your AI agent a simple thing.

"Find me the best monitor for my laptop. Under 15,000. Good for coding, easy on the eyes."

Easy errand. The agent goes off to do it. First stop, obviously, Amazon.

And it hits a wall. A CAPTCHA, a "sorry, something went wrong" page, or just a quiet block. The site has figured out this isn't a person shopping. It's a bot, coming from a data centre. So the door stays shut.

Fine, the agent tries other sites. It finds monitors. But the prices are in dollars. Half the models aren't even sold in India. The "fast delivery" it's excited about is fast delivery to somewhere in Virginia.

The agent isn't dumb. It just isn't here. It's sitting on a virtual machine in another country, seeing the internet from over there, and making decisions for me from over there.

That's the thought I keep coming back to. My agent has a brain. What it doesn't have is a home.

The agent is smart. The internet doesn't trust it.

Personal AI agents are seriously powerful now. They can browse, compare, fill forms, book things, do research for hours without getting bored. I wrote about where I think this goes in an earlier post, The next big shift: when personal AI agents start running our lives.

But the monitor story shows two problems nobody talks about much.

Problem one: some doors are shut. A lot of websites don't want bots walking around. CAPTCHAs, bot walls, "unusual traffic from your network" pages. And some of the biggest platforms, Amazon included, have made it pretty clear they'd rather not have outside AI agents shopping on their sites. You can understand why. But it means my agent, doing a perfectly normal errand for me, gets treated like a scraper.

Problem two: the agent is in the wrong place. This one is sneakier.

The internet was built to change what it shows you based on where you are. Your Google results depend on your location. Your feed depends on your location. Prices, stock, delivery options, "near me" results, even which language a page loads in. All of it looks at where the request is coming from.

So when my agent browses from a server thousands of kilometres away, it's not seeing my internet. It's seeing a stranger's internet. And then it tries to pick a monitor for me based on that.

If an agent is supposed to think like me, it should at least see what I see.

What if the agent lived at home?

Here's the thought I can't shake.

What if my agent sat in a small box at my place, right next to my Wi-Fi router?

Now when it goes looking for that monitor, it's coming from my home connection. Same network as my phone and my laptop. It sees prices in rupees. It sees what's actually in stock in my city and what can reach my pin code by Thursday. It gets roughly the same search results my house gets.

And to most websites, it doesn't look like traffic from a data centre. It looks like what it is: a request from a regular home connection, on behalf of the regular person who lives there.

There are two ways to build this.

1. The simple version: a home proxy. The agent still thinks in the cloud, but all its browsing goes out through a small device in my house. The brain is far away, the eyes are at home.
2. The fun version: the whole agent at home. A small box running a small language model, fine-tuned just for agentic stuff: browsing, clicking, filling forms, following steps. It doesn't need to write poetry. It needs to get errands done. Bigger thinking can still go to the cloud when needed, but the everyday work happens right here.

Either way, the agent gets something it doesn't have today. A physical place in the world. A body, kind of.

Isn't this just a residential proxy?

Sort of. Residential proxies already exist, and people use them for all kinds of things, not all of them good.

The difference is whose home it is. This isn't renting some random person's IP to pretend to be someone else. It's my agent, on my connection, doing my errands. It's the same as me opening Amazon and comparing monitors, just without me sitting there scrolling for an hour.

That honesty matters. An agent that acts as you, from your home, for you, is a very different thing from a bot farm pretending to be a thousand people. And if a site still says no to agents outright, that's their call to make. The box is about letting my agent see my internet, not about breaking into someone else's.

Will Big Tech just do this better?

Probably, eventually. Apple already loves the idea of things running on your device. Google, OpenAI, Amazon, Meta, all of them want to own your personal agent. One day one of them might ship a beautiful little box that does exactly this, or the platforms might agree on proper rules for letting agents in.

But that's going to take a while. Right now the big players are still figuring out who lets whose agent in, and on what terms. Until that settles, there's a gap. And gaps like that are exactly where small, scrappy builds get to be useful.

I don't need to wait for the perfect version to start.

Maybe the box is already in your pocket

Here's a twist I keep thinking about.

Your phone is already a small computer that lives with you. It's on your home Wi-Fi, it knows your location, it has a camera, a mic, a battery. It's basically the box I'm describing, just busy doing other things.

So maybe I don't need new hardware at all. Maybe I can take a phone, maybe an old one lying in a drawer, and turn it into a full-time agent. Not a phone with an assistant app on it. A phone whose whole job is to be the agent.

And flip that around. If a small agent box at home can browse, shop, book and reply on my behalf, how many times a day do I really need to pick up my phone? Maybe this hardware slowly starts doing the phone's job. You stop tapping through apps and just tell your agent what you want.

I'm not sure which way it goes. A new box, a converted phone, or something that ends up replacing the phone. I'll play around with all of them and figure it out.

And then the box grows legs

Here's where my head goes when I think about it a bit longer.

Today it's a small case next to the router. It browses. It runs errands online.

Give it a camera and it can see your desk and tell you the monitor you bought is too big for it. Give it a speaker and it can talk to you. Give it wheels, then arms, and at some point we stop calling it a box.

We call it a robot.

Maybe that's the actual path. Not robots arriving fully formed from a lab one day, but agents slowly getting bodies, one small piece of hardware at a time. And the first piece is just a little home for the agent to live in.

So I'm going to build one

This idea has been sitting in my head long enough, so I'm going to try it. A small device for my own home, or an old phone turned into one. My agent, on my network, seeing my internet.

Will it work the way I picture it? I don't know yet. That's the fun part.

I'll keep you posted. If I build it, the next post will be about exactly how.