A new AI agent or coding tool seems to come out every week, and I try most of them. They’re good, and I’m usually impressed within the first ten minutes of using one, but somewhere around minute thirteen I start making a list of things I’d change. I want it wired into my stack, I want it to remember one specific thing about how I work, and I want it to handle a workflow that makes sense to nobody but me. None of that is a knock on the tool, since anyone selling software has to build for the median user. I’m building for one person, so I don’t have to.

Software for one

I’ve been thinking about this a lot since building DeployMill, which is the platform I use to deploy my own software and, conveniently, the thing that lets me build more of it. I’ve deployed 89 apps with it in a couple of months, and most of them aren’t useful. Some were experiments, some were ideas that went nowhere, and some I built because I could. What changed isn’t the quality of the ideas, it’s the cost of going from an idea to a running application. I can start something on Saturday morning, be using it by Sunday, and rewrite whatever I don’t like the following week.

When you’re the only user, nothing has to be general. There’s no onboarding, no settings for people who aren’t you, and no compromise between what you want and what the average person wants, so the software can be as specific to how you work as you’re willing to make it. That’s how I want to build an AI agent.

Ziggy

So I’m building my own, and I’m calling it Ziggy, after the handheld computer Al carried in Quantum Leap that could answer almost anything he asked it.

The goal isn’t to beat the leading coding agents. I want the baseline features they all have, plus the two dozen specific things I’ve thought of that will never make anyone’s roadmap, and I want to try ideas about how an agent should work, change the interface, and connect it to my own tools without waiting for someone else to decide whether any of that is worth building.

Saturday to Monday

I decided to build Ziggy on a Saturday morning, scoped out the initial features, broke them down, and handed the work to my team of agents. By Monday morning it was running online with a backend, a web client, and a native iPhone app that covers roughly 70% of what I use the leading coding agents for today, and I already have a backlog of about 30 things I want to add to it.

A few years ago this would have been a multi-month side project of the kind that stalls out around month two. Now the hard part isn’t building it, it’s deciding what you want it to do.

Build yours

There’s a lot of debate about which agent, model, or framework is going to win, and I’m more interested in what all of them let me build this week than in who ends up on top.

If the tool you use is almost right, change it, and if nobody has built the thing you want, build it for yourself. It doesn’t need a business model and it doesn’t have to be useful to anyone else. It will make you better, though, because every one of these projects has taught me a stack or a technique that I’ve since used at work or to improve my next project.

Ziggy runs on DeployMill, it’s built entirely for me, and it probably won’t make sense for anyone else, which is fine. I’ll share what I add as I go.

I’m not trying to build the best agent. I want mine.