Human in the Loop: A Design Principle

What Human in the Loop Means

Human in the loop (HITL) is a design philosophy, not a feature. It means building AI tools where humans remain the decision-makers—AI provides analysis and recommendations, but the attorney makes the call.

In practice, this involves:

**Human initiation.** The attorney directs the AI, not the reverse.

**Human review.** Outputs pass through attorney evaluation before use.

**Override capability.** Any AI recommendation can be modified or rejected.

**Clear accountability.** Professional responsibility stays with the attorney.

Why This Matters

Bar rules presume human accountability. Rules of Professional Conduct address competence, confidentiality, and supervision of work—all built around the premise that a licensed attorney is responsible for legal services.

AI tools operate within this framework. Whatever tools you use, you remain responsible for the work product. HITL design acknowledges this reality rather than obscuring it.

Levels of Oversight

Not every task requires the same level of review:

High oversight: Client communications, legal conclusions, filed documents. The attorney reviews every output.

Medium oversight: Document categorization, data extraction, citation checking. The attorney reviews samples and audits patterns.

Lower oversight: Arithmetic calculations, formatting, calendar entries. The attorney sets parameters and reviews exceptions.

Even at lower oversight levels, the attorney maintains control—establishing the rules, monitoring results, and overriding when needed.

How We Build

Glass Box products are designed around these principles. Every output shows how it was derived. Every recommendation can be overridden. Every interaction is logged. Our tools do specific, bounded tasks—they support attorney judgment rather than replacing it.

This isn't a constraint we're working around. It's how legal AI should work.