LAB 001

Can an AI agent understand an interior-design brief?

A source-linked requirement register can reveal omissions before concept work begins.

Planned

Hypothesis

A source-linked requirement register can reveal omissions before concept work begins.

Design or workflow problem

A source-linked requirement register can reveal omissions before concept work begins.

Setup, inputs and constraints

Use one synthetic brief with a budget change, a missing dimension and two conflicting requirements. Compare extraction with a manually prepared reference register.

Agent and tool architecture

Brief Agent → source-linked candidates → human review → approved context. No model writes.

Human approval points

A person approves inputs, proposed design choices and release of the candidate. No autonomous production edits or machine operation.

Result

Not run. This record defines the experiment; no performance result is claimed.

What failed / failure tests

Test invented dimensions, missed conflicts and an outdated budget promoted as current.

What changed

Baseline protocol. Revisions will be recorded alongside the run that motivated them.

What was learned

No findings recorded. The criteria below define what the experiment must establish.

Acceptance criteria

Every accepted requirement has a source; all seeded conflicts and unknowns remain visible. Record corrections and reviewer time.

How to read contribution labels ↗

Read the architecture ↗

Visit the Lab ↗

Start with one real workflow.

Map your studio workflow