Where Humans Belong in an Automated Business

We build automation for a living, so clients are sometimes surprised by how often we insist on a human in the loop. The surprise comes from a bad mental model: that automation is a slider from "manual" to "autonomous" and progress means pushing it right. The real skill is knowing which gates are load-bearing.

The gates we never remove

Anything that binds the company. Contracts, statements of work, pricing commitments. Our own SOW generator drafts beautifully — and nothing it drafts reaches a client until a human approves it. The cost of the gate is minutes. The cost of a bad commitment is the relationship.

First contact in a sensitive relationship. An AI can draft the note to your biggest account. A person should decide whether today is the day it sends.

Go-live itself. Every Player we deploy starts in rehearsal — running, logging, touching nothing real. A person promotes it to the live stage deliberately. Autonomy is granted, never assumed.

Judgment about people. Hiring screens, performance signals, anything where the output is a decision about a human being. AI can structure the information. The decision needs an owner with a name.

Checkpoints versus bottlenecks

A checkpoint is a gate where the human adds judgment the system lacks. A bottleneck is a gate where the human adds latency and ritual. The honest test: what percentage of items does the reviewer actually change? Above ten percent, keep the gate — it is doing work. Below one percent, the gate is theater; instrument it and consider retiring it.

Automate the work. Gate the commitments.

The slider model gets this backwards. Removing checkpoints is not maturity — knowing exactly which three you would never remove is. Businesses that automate everything and businesses that gate everything fail the same way: slowly, and then suddenly. The ones that thrive can name their load-bearing gates from memory.