A proposed protocol for measuring the human review time behind agent output
A Fractional Futurist post lays out a per-task method for logging the human time and agent spend behind each accepted agent result.
A post on Fractional Futurist (October 16, 2026), shared on X by @bibryam, sets out a protocol for measuring the human effort and agent spend each completed agent task actually consumes, at an agreed quality bar. If you're budgeting agent-saved hours, this is the number to collect first. The author says it is a proposed method, not results from a completed experiment.
The core rules:
- Register every eligible task before work starts, so you don't keep only the runs worth sharing.
- Treat the business task as the unit. A rejected patch plus two revisions is one task with three attempts, and a new chat doesn't reset the cost.
- Log active human time as briefing, supervision, review, rework and verification. Pause the timer while the agent runs unattended. A missed entry is missing data, not a zero-minute review.
- Report human minutes and agent spend per accepted task across all started tasks, and split agent-assisted completions from manual takeovers so a rescue doesn't look like successful delegation.
The post's synthetic example: three tasks, 45 human minutes and $6 in charges, two accepted, works out to 22.5 minutes and $3 per accepted task.
The author notes the ledger can't prove savings on its own. That takes randomised comparison, as in METR's 2025 study, where developers took 19% longer with AI while believing they were about 20% faster, per this summary. Review time is where the cost tends to hide: daily.dev cites Faros AI data showing task completion up 34% in 2026 while median review time grew five times.
The author hasn't reported results yet.
Sources
- How to Measure the Human Cost of Agent Work · Fractional Futurist
- The compiler was supposed to end programming too | daily.dev · daily.dev
- Experienced developers were slower with AI and believed they were faster · AI-first company · Paweł Szczęsny
- @bibryam: How to Measure the Human Cost of Agent Work · @bibryam