The short answer
Most founders automate the most annoying thing in their week. The thing worth automating is the most expensive one, and those are rarely the same. Annoying tasks are visible, so you notice them. Expensive failures are usually omissions, and omissions have no friction at all, so nothing draws your attention to them.
That single mistake is why a large number of people try AI agents for a month, conclude the technology is overhyped, and stop. The annoying task is annoying precisely because you notice it. The expensive one is expensive precisely because you do not.
What this looks like in practice
The founder who spends a weekend building something to tidy their inbox, while the deals that went quiet three weeks ago stay quiet.
The founder who automates their weekly report — tedious, forty minutes — while every first sales call happens with no research because the research would have to happen at 11pm.
The founder who builds a content generator, because content generation is the demo everyone shows, while enquiries from the ads they are paying for sit unanswered for a day and a half.
In each case the automation works. It just does not matter.
Annoyance and cost are barely correlated
Annoyance is about friction and repetition. Cost is about consequence. They come apart because the expensive failures are usually failures of omission — the follow-up that never went out, the signal nobody read, the invoice nobody chased. Omissions have no friction at all. Nothing happens, and nothing continues to happen, silently.
Meanwhile the tasks that irritate you are the ones you keep doing. They are visible, they are present, and they are usually not what is costing you.
The rule of thumb: if it makes you sigh, it is annoying. If you did not know it happened, it is expensive.
How to find the expensive one
Price the failures rather than the tasks. For each of these, work out a real number:
Deals that went quiet. Count the open conversations more than ten days without a touch. Multiply by your close rate, then by your average deal. That is the annual cost of one habit.
Enquiries that go unanswered or answered late. Take your monthly acquisition spend and multiply by the share of enquiries that never get a substantive reply. That fraction is money you already spent and did not collect. Whether five minutes beats sixty has not been independently measured; whether something answers at all has.
The business living in your head. Price the hire you have postponed because nothing is written down, plus whatever one wrong forecast has already cost you.
Invoices nobody chased. Total sitting past 45 days. It is usually larger than founders expect.
Customers who left quietly. Last year's churn, and the retention number your next raise gets priced against.
| Failure | How to price it | Typical shape |
|---|---|---|
| Deals that went quiet | Open conversations untouched past ten days x close rate x average deal | Large, invisible |
| Enquiries answered late or not at all | Monthly acquisition spend x share never given a substantive reply | Money already spent |
| The business living in your head | The hire you postponed, plus one wrong forecast | Slow-burning |
| Invoices nobody chased | Total sitting past 45 days | Bigger than founders expect |
| Customers who left quietly | Last year's churn, priced at your next raise's multiple | Compounding |
Pick the biggest number. It will often be the least interesting item on the list, and that is the point.
Then build one thing
Not three. Three simultaneous agents means three mediocre ones and no clear read on whether any of them worked. Build the one with the biggest number, run it for two weeks, and check whether that number moved.
If you cannot say which number should move, you have not found the bottleneck yet. That is not a tooling problem and no amount of better prompting fixes it.
The uncomfortable part
Doing this often reveals that your most expensive problem is not automatable at all. It is a positioning problem, or a pricing problem, or you are selling to people who cannot buy.
That is a useful outcome. It is a much better month than one spent building something elegant that saves you forty minutes.