← Insights Trust

Where AI stops and a human takes over

Cody Reppert · Field notes

The most common question I get is not "what can AI do." It's "what happens when it's wrong." That is the right question, and firms deserve a real answer before they hand anything over.

Every system I build has a line in it, drawn on purpose, between what AI is allowed to do on its own and what always routes to a person first. That line is not a limitation of the technology. It is the design.

What AI handles

The repetitive, high-volume, low-judgment work: drafting, summarizing, formatting, moving information between systems, flagging what needs attention. The parts of the job that are draining precisely because they don't require a skilled person's judgment, and shouldn't be consuming it.

What a person is always responsible for

This is what "human-in-the-loop" actually means in practice. Not a disclaimer at the bottom of a page, a specific design decision about which steps a person reviews before anything moves forward.

Why this matters more than raw capability

A system that is impressive in a demo and unsupervised in production is a liability, not a feature. The question that actually matters is not how capable the model is. It's whether someone thought carefully about where it can act freely and where it can't, and built the system to respect that line every time, not just when someone remembers to check.

I do not ship black boxes and walk away. Everything I build stays maintained, and a person reviews anything that carries risk. That is not a slower way to get the benefit of AI. It is the only way that holds up over time.

If you want to see exactly where that line would sit for your firm, an AI Efficiency Audit is where we work that out.

Request an AI Efficiency Audit