A piece can look good and still be wrong for the brand.
Judgment asks why, one value at a time.
TRY THE JUDGMENT BOARD
ACTUAL BOARD EXAMPLES / AI-GENERATEDINDEPENDENT EXPERIMENT. NIKE IS AN EXAMPLE BASED ON AN EDITORIAL READING OF PUBLIC SOURCES; NO AFFILIATION OR ENDORSEMENT. SAMPLE IMAGES ARE AI-GENERATED. SAMPLE SCORES, DECISIONS, AND HISTORY ARE ILLUSTRATIVE.
AI made it easy to make more work. Reviewing it didn’t.
A team that once explored five directions can now make fifty. Producing another variation takes seconds. Deciding whether any of them actually belong in the world still takes people, attention and judgment.
That was the bottleneck I wanted to explore. I started with a practical question: could AI tell whether something was on brand? If it could, maybe it could help with that first layer of review.
But “on brand” became a lot less binary. A piece can follow every guideline and still miss the brand. Another can break several rules and completely nail it. So the question changed: could AI look beyond compliance and evaluate whether the work aligned with the values and intent behind the brand?
Judgment, made visible.
A piece can look good and still be wrong for the brand. This board asks why, one value at a time. Choose a brand, give it the values to work from, and test a piece. The agent shows its reasoning; you decide whether it has a point.
Whose values?
Values are always checked. Latitude only sets how far the look may travel.The Nike values are an editorial interpretation of public sources. The example images and scores are illustrative.
Brand values
Bring your own brand, judge more than one piece.
One image and its own copy per pieceAdd up to five images. Write a tagline, caption, or context for each one. You can add a text-only piece too.
Test your own work
Describe a pitch, caption or concept and add an image if you have one. The agent reports only what it can support; anything it cannot see comes back as unknown.
The four sample decisions above show how it works.
Pieces to review
Review one piece
Agent first pass, then a person decidesDecision comparison
One value at a time, across all workEach symbol is a value check for one judged piece. Select a row to open the full reasoning and human decision above.
Drift across all work
One odd choice may be an experiment. Three in a row might be a new house style nobody agreed to. Each dot is one piece: right means stronger value alignment; up means greater visual departure. Diamonds average the latest three. Drag a dot to explore a what-if route; the change is not saved.
Does it hold up?
Eighteen cases give the judge the same assignment. I made a draft call on each; change it when you disagree. The text cases test value logic, and four image cases test whether it can read a picture. Your disagreement is part of the test.
SCROLL WITHIN THE BOARD TO SEE THE FULL TOOL ↓
The agent shows its reasoning. You decide.
Judgment evaluates each piece against the brand values one at a time. For each value, the agent can return Pass, Partial, Fail, or Unknown, along with the reasoning behind its assessment.
Unknown matters. If the system can’t support a conclusion from what it can actually see, I’d rather have it say so than construct a beautifully worded explanation of something it made up.
You can use the Nike example or bring your own brand values and work, including images, copy and text-only concepts. The agent makes a first pass. A person decides.
Values on one axis. Visual departure on the other.
A piece can move pretty far from the house style and still strongly express what the brand stands for. It can also look extremely familiar while saying almost nothing.
Judgment separates value alignment from visual departure. Values are always checked. Latitude only sets how far the look may travel.
One unusual piece might simply be an experiment. If the work keeps moving in the same direction, something else may be happening. Judgment surfaces that pattern. Then a person can tighten the brief, flag the direction for discussion, or record that the change is deliberate.
Drift across all work
Four illustrative samples. The diamond is the average of the latest three.
↑ MORE VISUAL DEPARTURE · FOUR PIECES SHOWN / LATEST THREE AVERAGED
Eighteen cases put the agent to the test.
I made eighteen cases and gave the agent the same assignment across all of them. The cases test how it reasons about brand values, and how it handles uncertainty when it cannot actually read what’s happening in a picture.
I made a draft call on each first. Then the agent got its turn. You can run a case once or three times, which makes another problem visible: does the agent reach the same conclusion when you ask it the same question again?
The goal isn’t to make it agree with me eighteen times. The useful part is seeing where the reasoning holds up, where it changes between runs, and where another human looks at the result and says, no, I don’t buy that.
It works with the other tools. It solves a different problem.
Judgment isn’t a replacement for the tools that generate creative work or the systems that enforce brand standards. It sits after them, where people have to decide what the work means and whether it deserves to go out.
Create the work.
Check the rules.
Examine the meaning.
Make the call.
I started with a binary question. I ended up with a better one.
I started by asking whether AI could recognize if something was on brand. What I found is that “on brand” is doing a lot of work. It’s guidelines, it’s values, and it’s taste. What I want from AI isn’t a verdict. I want its reasoning. Then I want a person to decide.
Once those agreements, overrides and intentional departures start accumulating, you have a record of how people actually interpret the brand when real work is in front of them. That’s where this project connects to a larger question for me: how do we use AI to help evaluate taste without giving up our own?
Judgment is rare.