$20, $100, $200 a month, and nothing to show for it

I paid for my first AI subscription the way I pay for everything easy: without thinking. Twenty dollars. I make that back in an hour. The plan was fine, the limit was fine, and for a few weeks it felt like holding a tool that made everything possible.

Then the limit started to bite and I upgraded. Fifty a month. Then the fifty started to bite and I looked at the hundred and did the arithmetic in a way that made it look reasonable, which it does not if you read the number out loud. A hundred dollars a month is a phone bill and a gym membership and a decent dinner, and I was spending it on a chat window.

I did not feel it. That is the part I want to write down, because the feeling is the whole problem. Twenty dollars does not register. Fifty does not register if you have a job. A hundred does not register until you list it next to what else a hundred dollars buys, and by then you have been paying it for months.

The math I avoided

Twenty dollars is a bit over five hundred thousand dong. I know this now. For months I did not, because the number on the card was in dollars and the number in my head was "cheap," and those two things never met.

Five hundred thousand dong is not nothing where I live. It is a week of eating out. It is a decent mechanical keyboard, once, that lasts years. It is a phone data plan. It is a month of music and video and still change left over. I know exactly what else it buys, because I have bought all of those things, and every one of them is still here.

What I was buying instead was different in kind: a subscription I used for a few hours, forgot about, and paid for again.

The conversion I ran in my head instead was different and dishonest. I told myself the subscription was an investment in the thing I was building. Build now, the users come later, and the users are the reward. That sentence is true in a company with customers and a deadline. It is a story in a bedroom at midnight with an empty repo.

What I was actually buying

The expensive research on this is very good and I read it after the fact, which is the wrong order. There is a name for the trap I fell into and the name is "the gym trap": you buy the full membership, you go twice a week, and the maths never closes. Nobody sells you a two-visits-a-week plan, so you buy the plan for someone who lives at the gym and pay for capacity you never touch.

The other honest framing I found: I was not paying for the model. I was paying for the feeling of having access to it. The subscription is a door that stays open, and an open door feels like progress even when you never walk through it. A person I read described keeping a service for three months after they stopped using it because cancelling "felt like closing a door." I have done exactly that. Closing a door is not a reason to keep paying rent on it.

There is a third thing, and it is the one that stung. The bottleneck was never the AI. I hit the limits on the plan and upgraded, but the limits I actually hit were mine. I am the one who has to decide what to build. I am the one who has to work out what the feature should do, read the docs, test it, change my mind. That takes days. On the days I am thinking, I am not spending my quota. I am paying for a machine that sits idle while I do the slow human part, and then I pay for it again on the days I generate a thousand lines and delete them the next morning.

The results column was empty

Here is the accounting I eventually did, and it was not a spreadsheet, it was a feeling that had been waiting for me to sit still.

I shipped nothing that anyone uses. I have a folder of projects that got to the interesting part and stopped, which is where they always stop, because the interesting part is the build and the boring part is the distribution, and I only ever paid for the build. I have settings tuned to a model I will not use next month and prompts written for an interface I will leave. I have spent a year and a real amount of money and the honest output, the thing that exists in the world because of it, is close to nothing.

And the money is gone the way a subscription is gone: completely, with nothing to show. The keyboard is still on my desk. The books are still on the shelf. The meals were eaten and I remember them. The subscription was used for a few hours and then it was just a line on a card, and the line kept coming back.

The zero is not only the money. That is what I want to be clear about, because "I wasted some dollars" is a small confession and the actual loss is bigger. The loss is the year. The subscription kept the door open and the open door let me keep telling a story about building, so I never had to face the fact that I was not building, I was paying to not decide.

What actually helped

Two changes, and neither was a cheaper plan.

The first was to stop treating usage as the goal. For a while I measured a good evening by how much I burned, which is a terrible way to measure an evening and also exactly what the flat-rate pricing is designed to encourage. Now I measure by whether a thing exists at the end that did not exist at the start. Some evenings that is a small thing and the quota barely moved, and that is a fine evening.

The second was to be honest about which kind of work I was doing. Frontier models for the decisions that matter, cheap models for the boilerplate, and nothing at all for the days I am only thinking. The research calls this a split between tasks where quality changes the outcome and tasks where it does not, and the hard part is not the split, it is admitting which pile my evening actually falls into.

I cancelled one subscription and downgraded another. The world did not end and the projects did not speed up, which told me the subscription was never the constraint. That is a boring conclusion and I needed the invoice to reach it, which is a worse way to learn it than reading this would have been.

The embarrassing part is not the money. It is that the story I told myself, build now and the users are the reward later, was a story I never tested. I did not build now. I bought now, and then I told the story about building, and the two felt the same from inside the room. They are not the same and I found out the way everyone finds out, by sitting down with the number and having nothing to put in the other column.

If you are paying for a plan and the results column is zero, the plan is not the problem. The plan was never the problem. It just made the problem feel like progress, and that is what makes it worth writing down. Twenty dollars is cheap. Twenty dollars a month for a year, to avoid facing that you have not built anything, is the most expensive thing I have bought.