Set a retry budget: three attempts, each changing one thing

The task came back wrong. You ask again, a little more firmly. It comes back wrong in a different way. You ask once more. Forty minutes and six runs later, you have learned very little.

Retrying out of habit feels like progress and is usually just spending. Every run costs something: time, attention, and often money. A retry budget is a rule you set before the first run, so that the sixth attempt never happens by accident.

Why decide in advance

Mid-task, stopping is hard. Each failed attempt makes the next one feel more justified, because you have already put so much in. That is the sunk-cost trap, and it is nearly impossible to resist in the moment.

A number you chose while calm does the resisting for you. It also changes how you work: if you only have three attempts, you spend the first one thinking about what to change.

The rule

Set the number before you start. Three is a sensible default. Use fewer for expensive runs and more for quick, cheap ones.
Change exactly one thing per attempt. If you rewrite the brief, swap the input and change the tool all at once, you will not know which change helped.
Predict, then run. Write one line first: "I expect this to fix X." A failed prediction teaches you something a bare failure does not.
Log each attempt in one line. What you changed, what you saw. Three lines are enough.
When the budget is spent, stop retrying. Pick an exit instead: split the task, hand the hard part to something else, do that part yourself, or drop it.

A menu of single changes

If you are not sure what to change, work down this list, one item per attempt:

State the goal in a sentence, not just the steps
Add one example of a result you would accept
Narrow the scope to a single item or section
Supply input that was missing or unreadable
Reorder the steps
Use a different tool or model for this step
Shrink the job and run it on one item first

Illustrative example

This is a hypothetical scene, not a report of a real case. A researcher asks for a one-page comparison of the notice periods in three contracts.

Attempt one mixes up the clauses. Change: she adds the clause numbers to the brief. Attempt two is right for two contracts, and the third is missing. Change: the scan of the third was unreadable, so she pastes its text. Attempt three is correct. Her log is three lines long.

On another task with the same budget, attempt three is still wrong. She does not run a fourth. She does the troublesome part herself in five minutes, and moves on, with a log that tells her exactly what did not work.

When to extend the budget

Sometimes a fourth attempt is right. Extend only for a reason you can write down: "attempt three showed me the helper misreads dates, so one more run with the format stated." Do not extend on hope, and do not extend twice. The budget exists to make you decide deliberately, not to forbid being flexible.

Compare the retry with the alternative

Before each attempt, ask what it would cost to do the step yourself, and what the retry will cost in minutes and in money. A retry that costs more than the step it replaces is a signal to take the exit.

If your tool shows cost, read it before every run, not only after. If it does not, jot down the clock time instead. Either way, the point is to make the price of an attempt visible, because invisible costs are the ones that quietly accumulate.

A quick card

Number of attempts decided before the first run
One change per attempt, written down
A prediction before each run
An exit chosen in advance: split, hand off, do it myself or drop it
Cost or time noted for every attempt

Octuo, our personal assistant for work across AI, tools and specialist services, previews cost before and reports it after every turn. Octuo is available for macOS.