Review Offset
The time a generative tool saves returning as the time required to check what it produced, which is real, measurable in principle, and absent from every headline figure.
Reviewed July 17, 2026Stable
When not to use it
- Where output requires no checking, which makes the headline saving the net saving.
- Where review would have been performed anyway on human-produced work, since the offset is the additional review the tool created.
- As an argument that savings are illusory, when a majority reporting partial offset also report a net gain.
Reach for something else instead
- Timing both activities — the direct measurement, rarely performed because review is not treated as part of the task.
- Sampled review — checking a proportion rather than everything, which recovers time in exchange for a known and stated error rate.
Read more on the blog
- 72 seconds, or 30 minutes, and both are trialsTerritory 11 opens on the first subject this corpus has examined where the evidence is genuinely good. Registered trials, CONSORT-AI reporting, peer review, and effect sizes that still differ by a factor of twenty-five.
- Refactoring fell from 25% to 3.8%Survey evidence says AI improves code quality. Repository telemetry says the opposite. They are measuring different things, and the gap between them is where the productivity went.
- Lesson planning is 60 to 99% of the usageTeachers report saving between 2.9 and 14 hours a week with AI, every figure is self-reported, and the platform data shows they use it for planning rather than for the admin the case rests on.
Further reading
- Wong et al. (2021), External Validation of a Widely Implemented Proprietary Sepsis Prediction Model — burden created by a system and absorbed downstream by clinicians.
- METR (2025), randomised trial of experienced developers on their own repositories — measured time against estimated time on tasks involving generated output.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Quoting a saving without asking whether review is mandatory in that setting.
- Assuming the offset is captured in a self-report, when the respondent may not be the reviewer.
- Treating review as overhead rather than as part of the task, which is what removes it from the measurement.
At a glance
FieldApplied
Two questionsis review mandatory, and who performs it
Underreported systematically, not randomly
Where the reviewer differs from the author, no single respondent can state the net
Measurementtime both activities on one task
DifficultyBeginner