I built the same agent twice

I Built the Same AI Agent Twice Without Knowing

Key Takeaways

An AI agent inventory takes minutes to build and reliably turns up work you had forgotten you owned, including duplicates.

I audited around twenty projects and six or seven were dead, half-built or built twice.

A list tells you what exists. An operating system tells you what to reach for on a given day, which is the column that makes it useful.

The payoff is portability. When something new ships, the people who know what they own move their existing work into it.

I built the same AI agent twice. Months apart. No idea I had already made it.

This is not a tool problem, and that matters, because people reach for a new platform when the real mess is inside the one they already have. Agents. Skills. Projects. Prompts written months ago and never opened again.

Good work, all of it. Real, useful work I had built and tested. Sitting behind a wall where I could not see any of it, so I kept building new versions of things I already owned.

An asset you forget you have is not an asset. It is a file.

This post covers where the mess hides, how to inventory everything in minutes, and why this matters considerably more than a tidy folder.

Why Good Work Goes Invisible

An AI agent inventory is a single list of every agent, skill and project you have built, what each one does, and what state it is in, kept somewhere you look regularly.

The reason it is needed is that none of those things are visible by default. A skill inside Claude works beautifully and is almost impossible to see. They are not in a list in front of you. They are not on a dashboard. They sit behind a wall, and you only reach for them if you happen to remember they exist.

So I had built genuinely good skills, tested them, refined them, and then they went quiet. Out of sight. Not front and centre where I would use them.

My projects were worse. When I finally sat down and audited them, I had around twenty. Six or seven were dead, half-built, or things I had made twice.

Half-built things I had walked away from mid-build when something else came up. A few that no longer worked at all and I had never gone back to fix. And I could not have told you which was which without opening every single one.

That is the part that gets you. The work was not bad. I had no idea what state any of it was in.

Think about a car. If you do not know what is under the hood, you are driving blind, with no map and no guide, and you will go and buy a part you already have sitting in the trunk.

That is what happens to most people building with AI right now. Not carelessness. Volume, arriving faster than any way of seeing it.

The Scan That Takes Minutes

I had Claude inventory its own skills, and the whole thing took minutes.

I asked it to scan everything I had, list it out, describe what each one does, and push it straight into Notion.

What came back went further than a list. It was an operating system: what I have, what each one is for, and when to reach for it. That last part is the whole point, because a list tells you what exists while an operating system tells you what to use on a Tuesday when you have a task in front of you.

Then I did the same for projects. Full audit across all twenty. What is live, what is half-finished, what is dead. Cleaned all of it up, and that one took real time. I will not pretend otherwise.

Here is the part worth copying. I now re-run that same scan monthly, or weekly when I have been building a lot. It updates the inventory and rebuilds the operating system, so the mess cannot stack up again.

The cleanup is a one-time job. The scan is the habit. Those are two different things and people confuse them, which is why the first cleanup so often turns out to be the only one.

If you have an assistant or somebody on your team, hand this over. It is a clean delegation. They run the scan, keep the inventory current, and flag anything that looks abandoned. You review it.

The Real Payoff Is Portability

Something new ships every few months now, and each time it does, the people who know what they own get to move while everybody else starts over.

Claude Code arrived. Then Cowork. Something else is coming that neither of us has heard of yet.

Each time one lands, there is a question worth answering fast: can I take the work I have already done and use it here? If your material is organised and sitting in one place you can point at, the answer is yes. You carry it over and adapt it. If it is scattered behind walls in six places, you rebuild from scratch, doing the same work a second time because you could not find the first one.

You cannot move what you cannot see.

Here is what that looked like in practice. I worked out that skills are the way to hand work between Claude and Canva, automating that whole handoff. I already had a pile of agents writing social content, already built, already sounding like me.

ey were inventoried, I looked at the list and immediately saw which ones belonged in that new setup. I retweaked those, and the handoff was running. I built nothing new. I took what I had and pointed it somewhere better.

Had those agents still been invisible, I would have sat down and written them from nothing. And I would have thought I was being productive.

What To Do Next

One pass and you will find duplicates. I found six or seven in twenty projects, and I do this for a living.

Have Claude scan and list everything. Every skill, every project, every agent. Do not attempt this by hand, because the hand version is what you have been avoiding for months.

Send the output somewhere you will actually look. Notion, a Drive folder, whatever you already open daily. One place.

Add two columns: what each one is for, and what state it is in. Live, half-built, or dead. Three states is enough.

Set the rescan. Monthly, or weekly if you are building fast. Or hand it to your assistant and make it part of their week.

Start with the scan today. Everything after that is sorting, and sorting is easier than remembering.

Know What Is Under the Hood

I built the same thing twice because I could not see what I owned, and one scan plus a monthly habit fixed it permanently.

An AI agent inventory is not about being organised for its own sake. It is what lets you be quick when something new drops, and quick is where the advantage is right now.

You probably have more built than you think. Go and find out what you have got.

Frequently Asked Questions

Q: How do I see all the skills and agents I have built in Claude?

A: Ask Claude to scan and list them, describing what each one does, then send that output to a document or database you open regularly. They are not displayed in a single view by default, which is why they go unused.

Q: What should an AI agent inventory actually contain?

A: Four columns cover it: the name, what it does, when to reach for it, and what state it is in. The state column matters most, because half-built work is the material most likely to be rebuilt by accident.

Q: How often should I audit my AI tools?

A: Monthly for most people, weekly if you are building a lot. The first cleanup takes real time, and every scan after that takes minutes.

Q: Can I delegate this to a virtual assistant?

A: Yes, and it is one of the better tasks to hand over. Running the scan, updating the inventory and flagging abandoned work needs no decision-making from you until the review stage.

Q: How much duplicate work do people usually find?

A: More than they expect. My own audit covered around twenty projects and turned up six or seven that were dead, half-built or built twice, and I build these professionally. Beyond the wasted time, duplicates mean improvements get made to one version while you keep using the other.

Search
Categories