Design Assay: Manifesto#

The shift#

Science just learned that one person with agents can coordinate the work of many labs. A single founder can direct dozens of parallel experiments, screen tens of thousands of candidates, and keep only the few that beat the benchmark.

Design has not caught up. Most design is still made the way it was in 2015: a few people produce a few options, debate them in a room, and ship the one that won the argument.

Design Assay replaces the argument with an assay.

The principle#

Generate widely. Measure everything. Keep what survives.

1. Variation is cheap now. Agents can produce thousands of credible variants of a screen, a component, or a brand system in the time a team produces three. The scarce resource is no longer options. It is judgment at scale.
2. Judgment becomes a fitness function. Most of what a senior designer checks can be measured: accessibility, contrast, token fidelity, rhythm, performance, responsiveness. What cannot be measured is made explicit, scored by rubric, and logged.
3. Selection is the craft. The designer's job moves from drawing the answer to defining what "fit" means, then steering selection. Taste is not removed. It is encoded, made repeatable, and applied ten thousand times.
4. Receipts or it did not happen. Every assay publishes its numbers: variants generated, hits, hit rate, gain over baseline, time. Claims without run records are opinions.

Why one person#

A studio scales by adding people. An assay scales by adding compute. One systems designer with a clear protocol, a local inference fleet, and frontier models for judgment can run what used to require a team of twenty, and run it with more rigor, because every decision leaves a record.

Research basis#

The idea that creativity works through variation and selection is not new. Donald T. Campbell described creative thought as blind variation and selective retention (1960). Dean Keith Simonton built a body of research on this model of creativity. Design Assay takes that theory and makes it an engineering practice: the variation is generated, the selection is measured, and the retention is versioned.

What we promise#

• The protocol is open.
• The fitness functions are public and versioned.
• Every published result has a run record anyone can inspect.
• Humans stay in the loop at the points where judgment cannot yet be measured, and those decisions are logged with reasons.