A few months ago, I kept running into the same frustrating loop.
I would ask my agent to help with something: plan a trip, organize a training program, think through nutrition, make a dashboard. Eventually, the work would need to happen inside some other product. That meant learning its UI, using its built-in agent, or asking my agent to drive a browser.
None of those options felt right.
I don't want a different mediocre agent for every app. And even when an app's agent is good, it doesn't have the context of my agent: what I'm working on, how I make decisions, or what happened in the conversation five minutes ago. Browser automation can bridge the gap, but it is slow and brittle. It can turn a simple request into a massive token faff.
What I really wanted was much simpler: tell my agent what I need, let it make the thing, and give me a useful interface when an interface is actually helpful.
That is the idea behind Charming: a cloud for small personal apps you build with your AI agent.
From apps with agents to agents with apps
Until recently, the default model was simply to use an app. Now, as assistants become part of how we work, many apps are adding an assistant of their own.
I think that model is wrong. People increasingly want to use their agent with their apps. Their agent is becoming the place where their context lives. The apps around it should be capable tools and interfaces, not isolated little worlds that demand to be learned from scratch.
Sometimes a chat response is perfect. Sometimes you need a trip plan your partner can edit, a dashboard your team can use, a form that captures information cleanly, or a view of your workout history. Those are apps. But they should feel like a natural extension of the work you are already doing with an agent.
The important point is not just that these apps do not need to become venture-scale companies to matter. For a lot of software, the right shape may be an app that is completely yours: built by your agent around exactly what you need, then changed whenever you need it to change.
A meal planner for a family, a tracker for a team, or a tool for one oddly specific need can be enormously useful to the people who depend on it. Today, we generalize software because it is expensive to make a different version for everyone. As the cost of making software falls, more software can be personal by default.
Personal apps are a different category
We already know how to build and host software. If you're technical, there are plenty of good choices: GitHub, modern hosting platforms, and every cloud provider under the sun.
But those tools are largely shaped by the needs of production software: flexibility, control, scale, complex authentication, and the possibility that millions of people will use the same product in different ways. Those are the right constraints for a startup. They are often the wrong constraints for a personal app.
A personal app is usually much more specific. It can have exactly the data model, APIs, and interface one person or small group needs. There are no years of migrations to protect. There may be no anonymous users. The people using it are often trusted: a partner, friends, coworkers, a small team.
Generalized apps will still matter, especially when people need to participate in the same shared system. But they do not need to be the default answer to every problem. An agent may eventually help the personalized tools we each use work together, without asking everyone to adopt the same interface first.
That changes what the platform should optimize for.
It should make identity and sharing easy. It should make the app reliable enough that you trust it when you need it. It should handle the boring operational parts without requiring someone to become an expert in deployment, auth, and infrastructure. And it should leave enough room for an agent to make changes as needs evolve.
In some cases, that may mean letting collaborators edit the app itself. In others, it may mean letting someone fork an app and make it their own. Those are unusual ideas in traditional software, but they make a lot of sense when the app is a living tool for a small group.
"But aren't people just building toys?"
Right now, a lot of what people build with AI looks like a toy: small games, one-off dashboards, trackers, and prototypes. I don't see that as evidence that this is small. I see it as evidence that the medium is early.
Early personal computers were also full of little, idiosyncratic programs. The people making them were learning a new skill and finding uses nobody had planned for yet. That is what is happening now as more people learn to make software with an agent.
Not everyone who can write becomes a bestselling novelist. But everyone writes. They write notes, policies, plans, messages, spreadsheets, and documents that coordinate real work. Some of the most consequential writing in the world is read by only a handful of people.
I think apps will work the same way.
Not everyone will build a billion-dollar company. But many more people will build software for their family, their team, their community, or a particular problem only they have. Some of those apps will be ephemeral. Others will quietly run workflows that matter a great deal. The important thing is that making them becomes normal.
There will be templates, shared patterns, and better ways to teach people how to make useful interfaces. But the real unlock is not a library of generic apps. It is the ability to start with a useful shape and then make it exactly yours.
The bet
My bet is that personal software becomes a normal part of how people think and work. Your agent will help you create and maintain a growing set of small tools, and those tools will be more personal, more adaptable, and more shareable than the software we use today.
Charming is my attempt to build the layer that makes that practical: a place where those apps can live, be used, be shared, and keep evolving.
We're early. That is the fun part. If you're building a personal app, or you have a workflow you wish existed, take a look at Charming. I'd love to hear what you make of it.