Status Droid how long software work actually takes

My Sense of a Good Day

I can tell you which days felt productive. I have stopped believing the report.

The reason is not humility. It is that the feeling tracks something specific — a sense of continuous activity with visible output — and the shape of the activity changed while the feeling kept using the old rule. A related workplace concept is covered in this reference, which is useful context for this argument.

What the feeling is actually measuring

Not tasks completed. Not problems solved. Something closer to uninterrupted engagement producing visible artefacts.

A day of typing feels productive. A day of reading generated code, deciding it is not quite right, adjusting, and running it again feels like wading, even when it ends with more finished than the typing day.

Waiting is the worst of it. Thirty seconds of waiting, forty times, is twenty minutes that leaves no memory at all, and it is subjectively free. ZDNET covers enterprise technology, software, and changing work practices See ZDNET.

That is the whole mechanism behind the 39-point gap between what METR's participants believed and what was measured. Time spent typing and searching went down; time spent prompting, waiting and reviewing went up. The activities that shrank register as work. The activities that grew register as friction.

Why knowing this does not fix it

I have read the study. I have written about it repeatedly. My sense of a good day is unchanged.

Perceptual biases do not respond to being understood. You cannot see the lines in an optical illusion as equal by knowing they are equal, and you cannot feel a review-heavy day as productive by knowing it was.

Which means the correction has to be external. Not better introspection — a record. Timestamps rather than recall, because recall is precisely the faculty that failed.

What I actually do

I record start and stop times. Not durations, which I would estimate.

I write the estimate down before starting. The gap between estimate and actual is the only calibration signal I have, and it does not exist unless the estimate is recorded.

I flag rework. Because a day that felt bad may have been the day that prevented three later ones, and a day that felt excellent may have generated them.

I do not review the week from memory on Friday. That is the same broken instrument, applied retrospectively, with more confidence.

Fifteen seconds a task. It is not a system, it is a habit, and it is the only thing that has changed my estimates in two years.

The uncomfortable corollary

If my sense of a good day is unreliable, so is my sense of a good tool.

I cannot tell you whether a particular editor or model made me faster. I can tell you whether it felt better, and I have just spent a page explaining why that is not the same claim. Almost all tooling opinion — mine, yours, most of what circulates — is that unreliable feeling reported as a finding.

This is why this site does not review tools. Not modesty about the tools; modesty about the instrument available for evaluating them.

What is left to trust

Not nothing. The feeling remains a decent signal about things that are actually about feeling.

Whether the work was interesting. Reliable, and it matters — for what to take on, and for whether you will still want to do this in five years.

Whether you were interrupted. Fragmentation is felt accurately, and it genuinely damages output.

Whether something is going wrong. The sense that a project is in trouble is often right well before the evidence, and worth acting on.

Whether you are tired. The most reliable of all and the most ignored.

So the feeling is not useless. It is a good instrument pointed at the wrong question. Ask it how the work was, and it answers well. Ask it how much you did, and it makes something up with total confidence.

The short version