In late September, I set up a team of AI agents to help with Downslope.
The idea was to hand off some of the work around the site and the textbooks I was publishing. One group of agents handled marketing research and drafts. Another handled steps in producing the books. A pair of agents acted as managers, writing daily summaries and weekly plans.
Within a few days, there were eleven agents and seven scheduled routines. Every one of them ran on the same AI usage I pay for to build Downslope itself.
For a while, the output looked productive. Reports arrived on schedule. Tasks moved across a board.
Then two things happened.
On October 10, four of the agents hit a usage limit and stopped. They sat in an error state for about a day before anyone noticed. Nothing else broke while they were down.
A few days later, one of the manager agents was still pushing another group to proofread the first textbook. The books had already been published. The work it kept assigning no longer needed to be done.
On October 14, I stopped all of them. They were using up too much of my AI usage, and much of that usage was going to work I did not need.
The agents were not the problem. They did exactly what they were set up to do. The problem was that they kept doing it after the reason had changed.
Recurring costs do not check in
Most household spending requires a decision each time. Groceries, gas, a new pair of shoes. Someone sees the price and decides.
Recurring costs work differently. The decision happens once, at signup, and then the cost repeats on its own. A streaming service. A meal kit. Cloud storage. A phone plan with more data than anyone uses. An app with a monthly fee. A membership started for a goal that has since changed.
Some costs are usage-based, which adds another layer. The cost grows with activity, not with value. Delivery orders, ride sharing, in-app purchases, and pay-as-you-go plans can climb quietly because each individual charge feels small.
None of these send a message saying the reason for them has gone away. A gym membership keeps billing after the schedule changes. A subscription for a child's activity keeps renewing after the season ends. A premium plan bought for one project keeps charging after the project is done.
The billing has a schedule. The review usually does not.
Give every recurring cost a review date
The step that would have saved me the most was simple. When I started the agents, I could have picked a date to ask whether they were still worth it.
A household can do the same with any recurring cost.
When signing up for something new, a review date on the calendar helps. Thirty or ninety days out works for most things. The question on that date is not whether the service is good. It is whether it did something useful for the household in the last month.
For costs already running, a short review a few times a year works well. List every recurring charge from the last two or three statements, including the small ones and the annual ones that are easy to forget. Then ask three questions about each.
The first question is what it actually did last month. Not what it could do. What it did.
The second question is whether the original reason still exists. The class, the project, the season, the goal.
The third question is whether it would be worth signing up again today, at today's price.
Consider a household that finds about $60 a month in recurring charges: two streaming services, cloud storage from an old phone, a fitness app, and a news subscription. One streaming service and the storage get used every week. The other three have not been opened since spring. Canceling those three saves $41 a month, close to $500 a year, without giving up anything the household was actually using.
For anything uncertain, pausing can work better than canceling. Many services allow a pause or a downgrade. If nobody notices for a month, the answer is usually clear. That was true of my agents too. Four of them stopped for a day, and nothing suffered.
Keep what earns its place
To be clear, this is not an argument against subscriptions or automation. I still use AI tools every day to build Downslope. Many recurring costs are worth every dollar, and some save more money than they cost.
The point is that recurring costs deserve the same attention as any other spending, on a different schedule. A one-time purchase gets one decision. A recurring cost needs that decision made again from time to time.
Stopping the agents was not a failure. Setting them up showed me what the work actually needed, and what it did not. That is a reasonable thing to pay for once. It is not something to keep paying for every month.
A helper that keeps working after the job is done is not really helping anymore.
Which of your recurring costs is still running for a reason that no longer exists?