I don’t shop for tools any more. I build them.

The parts of a hand tool laid out on yellow paper before assembly: a wooden handle blank with pencil marks, a steel head, a brass ferrule, two brass screws, a hex key and a paper template of the finished tool.

Finding the right tool used to mean searching, trialling a few, settling for the closest fit and paying for it every month. Sometimes the tool I wanted just didn’t exist. Now I build my own with AI, shaped to exactly how I work. Here’s how that changed, three ways to get started, and where building for an organisation is a bigger job.

The old way

You notice a gap. Tasks are slipping, or you can’t see where your projects stand. So you search “best tool for X”, read a few comparison posts that all recommend the same apps, and sign up for two or three trials.

One is close, but it wants you to work its way. Another has the feature you need on the next price tier up. A third does everything except the one thing you wanted. You settle for the closest one, pay the monthly fee, and use a fraction of what it does.

And sometimes there’s nothing. The tool you have in mind doesn’t exist, or only exists as part of something much bigger and more expensive.

What changed

With AI, building a small tool for yourself has become quicker than that search. You describe what you need, the AI builds a first version, you use it, and you ask for changes. Often it takes less time than the search used to.

The cost changes too. Instead of another subscription, it’s mostly your time and the AI tools you probably already pay for.

Three ways to start

  1. 1. Your own idea

    Sometimes you just know what you want. I keep several side projects going at once and kept losing track of which were active, parked, or waiting on a decision. So I built Cowork HQ: one page that shows every project, its status, the open questions and the history behind it.

    Cowork HQ: one page with every project, its status, the next step, open questions and the planning sessions behind it.
    Cowork HQ, shown with sample projects.

  2. 2. A specific need that nothing quite fits

    Some needs are too specific for any product. I wanted help planning my posts: what’s worth writing about, drafts ready to edit, and a weekly read of what worked. Content tools give you a calendar and a scheduler. What I needed was a set of AI agents working from a folder: one reads the news each morning and writes me a brief, one drafts next week’s posts, one reads my numbers on Monday. None of them can post anything. I do that.

    The same goes for checking AI-built work. Instead of paying for separate review tools, I wrote my own checkers as skills, for example a security review, a UI design review and an information-architecture review.

    (One thing I noticed along the way: some of what I thought were tools didn’t need a screen at all. The post planner is mostly files.)

    The growth folder: a briefs folder with one brief per morning, a queue folder of drafts, and this morning’s brief open.
    The growth folder: a brief each morning, drafts waiting in the queue. Sample content.

  3. 3. An existing product as inspiration

    You don’t have to start from a blank page. Earlier this month I had an AI agent rebuild a few popular apps to see how far it would get. Two of the results I kept: Tack-board, my own whiteboard, and Burrow, my own team chat. They’re not copies of anything I’d sell. They’re my versions, with what I use and nothing else. (How I did it →)

    Tack-board, a whiteboard with a simple flow and sticky notes, beside Burrow, a team chat with a builds channel open.
    Tack-board and Burrow, shown with sample content.

What you get

  • It fits. It does the job the way you work, not the way a product team guessed most people work.
  • Nothing extra. No menus full of features you’ll never open.
  • It changes when you do. New need next month? Ask for it.
  • No monthly fee, and your data stays where you put it.

From 1 person to 1,000

Of course, a tool for five people isn’t the same as one for a thousand. But the building part is the same: you describe it, the AI builds it, you use it and improve it.

What changes is everything around it. The more people use it, the more it needs: proper sign-in and permissions, security that holds up, their data kept safe and backed up, and someone looking after it next month and next year. That’s the real work at scale, and AI makes that much easier too.

So start with yourself. Then a small team. The step up is real, but it’s no longer out of reach.

Before you go looking for a tool

A first prompt to try:

I want a simple tool for my own work. The one job it does: [one sentence]. Inspired by: [your idea, or a product you like and what you’d change]. Who uses it: only me. What I need to see or do on screen: [list]. Keep it to a single page that runs in my browser and saves to [a file / my browser]. No sign-in, no server. Build the smallest version first, then ask me what’s missing.

If there’s a tool you keep paying for and barely use, that’s a good first one to try building yourself.