What is the autonomy tax?

The temporary cost of human-in-the-loop approval on agent decisions, paid in human time and delay until the team works out how to remove the checkpoint.

How does it work?

Each pause for approval costs the reviewer’s time, a delay in processing, and the load of switching context. A human checkpoint in every sensitive workflow removes the scale advantage but keeps the cost of building the agent. The reviewer sees only what the agent shows, so the approval often adds delay without real judgment.

Where does the human gate sit?

The essay moves the human from approving single decisions to setting boundaries. People decide what agents can and cannot do, resolve the real exceptions, and adjust boundaries from performance in aggregate. If nearly every recommendation is approved unchanged, the checkpoint adds latency, not judgment, and that is the signal to remove it.

What does it change for the team?

People act as architects and auditors, not gatekeepers. The move starts with low-stakes workflows where errors are easy to reverse, then raises the ceiling as performance holds. The key metric is recovery time, not accuracy.

Which essays cover it?

• Human-in-the-Loop Is a Temporary Placebo

@misc{couch2026whatistheautonomytax,
  author = {Couch, Chris},
  title = {What is the autonomy tax?},
  year = {2026},
  howpublished = {Lead to Cash},
  url = {https://chriscouch.ai/what-is/autonomy-tax/}
}