The Tools We Forgot We Were Renting

There's a specific kind of forgetting that only happens with money that leaves automatically.

I found mine in a spreadsheet. Ninety days of card statements, filtered for anything that smelled like software, sorted by amount. The number at the bottom was $300 a month, and I sat looking at it for longer than I'd like to admit, not because it was ruinous, but because I had no memory of agreeing to it.

Every line was defensible. That was the strange part. I could reconstruct the reasoning behind each individual subscription, and each reconstruction ended with yes, that was sensible. The sum of a long series of sensible decisions had become something I would never have chosen.

 

The economics of "that's nothing"

Software found the price point that switches off judgement. It sits somewhere around fifteen to thirty dollars a month below the cost of a dinner, above the cost of a coffee, in the exact band where a purchase feels too small to think about and too useful to skip.

Call it the nothing threshold. Below it, we don't evaluate. We just subscribe.

The threshold works because we experience price in the units it's quoted in. Nineteen dollars a month is a rounding error against a monthly budget. Two hundred and twenty-eight dollars a year is a thing you'd think about. It is, of course, the same money. The unit is doing all the persuading, and it's chosen by the seller.

What makes AI tools particularly good at this isn't that they're cheaper. It's that they arrive faster. A conventional software category produces a serious new competitor every few years. Since 2025, this one has produced several a month, each solving something slightly annoying, each priced comfortably below the threshold, each arriving at a moment when you happen to be annoyed by exactly that thing.

You don't build a stack. You accrete one.

 

The last-opened date

The most uncomfortable column in my spreadsheet wasn't price. It was the date I'd last actually opened each thing.

I had a story about my own tools a picture of myself moving fluently between them, each one earning its place. The dates said something else. Tools I'd have described as daily were monthly. Tools I'd have described as monthly, I hadn't touched since June.

This is worth sitting with, because it isn't really about software. We keep paying for the version of ourselves we intended to become. The transcription tool is for the person who was going to start recording interviews. The research assistant is for the person who was going to publish twice a week. The subscription doesn't lapse when the intention does, it just quietly continues, billing us monthly for an aspiration.

A subscription is a small, recurring vote of confidence in a future self. Most of them lose.

 

What I cut, and what it cost me

The instinct will be to make this triumphant. Resist it. The essay only works if the trade-offs are real: what you cut and why, the one thing you cut that you regretted, and what got worse what you kept without hesitation, and what it does that nothing else does.

The cutting was easier than I expected and less satisfying than I'd hoped. I went from $300 to $100 Nothing important broke. And the tools that survived didn't survive because they were the best in their category, they survived because they'd become load-bearing. They'd wound themselves into the way I work until removing them would have required rebuilding something.

That's the real moat in software, and it has almost nothing to do with features. It's not that a tool is good. It's that you've built around it.

 

The second inventory

Somewhere in the middle of the audit I stopped counting money and started counting access.

Every one of those subscriptions had asked for something at the beginning, in the polite way that consent screens ask: permission to read my email, to see my calendar, to reach into my files. An API key here. Authority to act on my behalf there. I had granted all of it, one screen at a time, always in a hurry, always at the moment when granting it was the only thing standing between me and the thing I'd come to do.

Nobody grants permissions thoughtfully, because permissions are always requested at the exact moment you are least interested in thinking.

And then I found the thing that changed the shape of the whole exercise: 30 of the tools I had already cancelled still held live access to my accounts.

I'd stopped paying them months ago. The billing relationship had ended. The trust relationship had not, because those are two different relationships and only one of them has a cancel button that's easy to find.

There is something genuinely unsettling about a list of applications that can read your correspondence, half of which you no longer recognise by name, some of which belong to companies that may no longer exist. Not a breach. Just an accumulation. The same accretion that built the spending, running quietly on a different axis.

 

Why this stops being a housekeeping problem

For most of software history, permission was a fairly abstract thing. An app could read your email. In practice it read your email when you were sitting there asking it to.

Autonomous agents dissolve that distinction, and they dissolve it deliberately, the entire value of an agent is that it acts without checking in. That's the product. That's what you're paying for.

Which means the mental model most of us are running, the one where I'll notice if something goes wrong, has quietly expired. You will not notice. Noticing was a side effect of being present, and being absent is the feature.

I don't think this argues for using fewer agents. I think it argues for a different habit: treating the permission as the purchase.

The question at the consent screen isn't do I want this tool. It's what am I handing over, for how long, and how would I ever take it back? Separate credentials for anything that acts on its own. Read-only until something has earned more. Rotation on a schedule rather than in a panic. And a hard line around anything with financial reach never a tool you're still trialling, never a credential you also use by hand. (That last part is educational, not financial advice, I'm describing my own hygiene, not telling you what to do with your money.)

One hour

None of this is complicated, which is exactly why it doesn't get done. Nothing forces it. No deadline arrives. The pile is not painful enough on any given Tuesday to be worth a Tuesday.

So: an hour. Card statement in one window, spreadsheet in the other. Price, converted to annual. Last-opened date. Then one question per row what breaks tomorrow if this vanishes tonight? and a willingness to accept a one-sentence answer or none at all.

Then the second list. Open the connected-apps page on your email, your calendar, your files. Read it all the way down. Revoke what you no longer use. Rotate what you've forgotten.

Most people find 100 of software they didn't know they were buying. Almost everyone finds at least one thing holding access they would never grant today.

The money is worth recovering. But the reason to do it is the other list, the one you've never read, that's been growing the entire time, one hurried yes at a time.

I write about AI tools, automation and the security side of both. The video walkthrough of this audit, plus the running cost teardowns, live on AiStackGuru.