Kerta Consulting
All posts

AI · Field Notes

If You Can Do It, an Agent Can Do It

Developers keep waiting for an MCP - the official integration that lets an agent use some tool. Most of the time you don't need it. If you can do the thing yourself on a computer, the agent can already do it as you. You just ask for the result.

24 Jun 20263 min read

Developers keep waiting for an MCP.

If you haven't met the term: an MCP is an official little adapter that lets an AI agent plug into some specific tool or service - a sanctioned doorway. There's a real protocol behind it, and for some jobs it genuinely helps. So a lot of developers see a thing they want their agent to do, find there's no MCP for it yet, and wait. Or wish one existed. They treat the missing integration as a locked door.

Here's what most of them are missing. The door isn't locked. If you can do the thing yourself - log in, click through, read the page, copy the number, fill the form, pull the file - then an agent can do it too, as you, with the same access you already have. You don't ask "is there an integration for this?" You ask for the result. The MCP is a nice white-tablecloth invitation to a room you can already walk into.

If you can do it yourself on a computer, the agent can already do it - as you. You just ask for the result.

I want to be precise, because this is the part people skip. Anything. If you can do it on a computer as yourself, you can hand it off. Anything you do with a mouse, a keyboard, a login, and some patience is automatable now - not someday, now. Most of the genuinely useful things I've automated this year had no integration, no API, no MCP. They had me, describing the result I wanted, and an agent doing what I would have done.

And it's worse for everyone who isn't a developer. They don't have the wrong mental model - they have no model. The idea that the boring computer task they do every week could just be done for them hasn't landed at all. Hand them a perfect integration and they still wouldn't reach for it, because they don't know there's anything to reach for.

This is the misunderstanding I keep running into, and it isn't small. Devs wait for permission that already exists. Non-devs don't know the door is there. In between sits an enormous amount of work - real work, the dull kind and the important kind - that could be handed off today and mostly isn't.

What gets to me is how this ends. Eventually it'll be obvious. Everyone will hand off the dull computer work without thinking about it, the way we look something up without thinking about it now. And when that happens, nobody will remember this moment - the years when the thing was already possible and almost no one used it. We're living in a blank the rear-view mirror won't even show, the same one I keep coming back to with the early internet. So I'll say it now, while it still sounds strange: if you can do it, an agent can do it. Stop waiting for the door to be built. Walk through the one that's open.