I logged every AI-assisted coding task for a month. Here is the honest ledger

I keep asking people to measure, so it is only fair that I did it myself. For September I logged every coding task at work where I used an AI assistant: what it was, how long it took, how long I estimate it would have taken without, and whether I had to fix the output. The estimates are mine and biased; I tried to be conservative in both directions.

The ledger

tasks logged:                       47
clearly faster:                     29   (est. 11.5 h saved)
about the same:                     11
slower (fixing confident errors):    7   (est. 4.0 h lost)
net:                                     about 7.5 h saved
time spent coding that month:            about 70 h

Roughly a 10% gain on coding time. Real, useful, and nowhere near 10x.

Where it saved time

• Boilerplate: argument parsing, plotting code, test fixtures, SQL for a schema I already understood. The biggest wins and the least interesting.
• Unfamiliar APIs: "how do I do X in library Y", where I could check the answer immediately by running it.
• Translation: a pandas snippet to SQL, a bash loop to Python.

Where it cost time

• Seven tasks where the answer looked right and was not: a deprecated parameter, an off-by-one in a date window, and a statistical test that assumed independence when my samples were paired. That last one cost the most, because the code ran fine and the p-value looked lovely.
• The pattern: it costs time when I cannot verify the output faster than I could have written it.

What I would actually conclude

• The gain is concentrated in tasks where checking is cheap. Choose tasks accordingly.
• Seniority matters in a direction people do not expect: the paired-samples error would have gone unnoticed by someone who did not already know the answer. The tool makes me faster partly because I can tell when it is wrong.
• I will keep using it. I will not claim it doubled my output, because it did not.

Jo, I believe you owe me a spreadsheet.