Five QR designs later, I rewrote the brief.
The AI produced five visual directions. I rejected the result because I needed something that belonged on a restaurant table.
this is only a design experiment, dont change any logic or code just design
Five different themes. Five different sets of colours and type. The assistant had done the obvious thing I asked for, and I still did not like the result.
The missing part was the object itself. This was going to sit on a restaurant table. I needed a guest to recognise it, understand the invitation, and know what to do with their phone. A treatment around a square was not enough.
See the original requestFrom the development record
Design the object around the scan.
A reading diagram of the physical brief. This is not a recovered print design or an application screenshot.
- 01
Notice it
The table object needs a clear visual invitation in its real setting.
- 02
Understand it
The person should know what scanning will open before reaching for a camera.
- 03
Reach the menu
The code and its destination must work together. The physical hierarchy does not end at an attractive border.
The next evidence would be a physical print and a scan test at realistic distance and lighting.
The original request31 Jan 2026 · excerpt
this is only a design experiment, dont change any logic or code just design
Excerpted from the development conversation. Punctuation is lightly edited; the surrounding story is an edited retrospective.
01
I separated the experiment from the application.
The existing QR designs felt too basic, so I asked the AI coding assistant for five polished prototypes with distinct themes, colours, and typography. I also set a boundary: create standalone HTML experiments and leave the product logic alone.
That was useful because the decision I wanted to make was visual. I did not need five competing implementations of ordering or table identification. I needed to compare how an invitation to scan could look and feel before choosing what deserved to enter the product.
The first response listed aurora, editorial ink, neon, terracotta, and glass treatments. It described them as static experiments and explicitly did not guarantee they were scannable. Those were the deliverables. A live ordering experience had not been changed.
02
Different styles were not the same as different answers.
“logo on the top and and the qr code and then text on the bottom to scan extra.”
My reaction was direct. I shared an example and explained the hierarchy I expected: identity at the top, QR in the middle, instructions below. The restaurant specific details in the example were not the point. The structure was.
This is a useful distinction when directing AI. A reference contains both the thing worth learning from and details that do not belong in the new result. If I only say “make it like this”, I leave the model to decide which is which.
So I clarified the physical format as well: A5 and A6. That changed the task from decorative QR treatments to printable layouts with a recognisable reading order. A narrow portrait sign has a top, a centre, and a bottom. The scan target must sit inside a complete invitation.
03
The next response was much closer to the actual brief.
The assistant then reported five new files, each with A5 and A6 artboards: Obsidian Precision, Atelier Warmth, Neon Pulse, Prism Glass, and Editorial Mono. Those were distinct visual directions within the layout I had clarified.
That is the turning point in this story. I did not get the result I wanted by adding more adjectives about polish. I changed what the model was supposed to design. The output now had a physical format and a structure against which it could be judged.
For the guest, that structure answers three questions in order: whose sign is this, where do I aim the camera, and what will opening it do? For me as the person building the product, it becomes a much clearer way to compare the alternatives.
04
A good looking QR is still only half the job.
A table sign connects paper, a camera, and the page that opens. Each can work in isolation while the whole experience feels awkward. A beautiful layout can crowd the code. A scan can succeed but open a page that gives no confidence it belongs to the same venue.
The practical review I would use now starts with the final object: print at the intended size, check the instruction at a realistic viewing distance, scan the actual encoded destination, and inspect the page it opens. Glare, lighting, paper finish, and the amount of surrounding decoration all deserve attention. These are next checks, not tests I am claiming happened in January.
DENSO WAVE’s guidance gives one concrete design constraint: a standard QR symbol needs an empty margin four modules wide on every side. A module is one small unit in its grid. That means the blank space is part of the functional object, not leftover room for a border or flourish.
DENSO WAVE: QR code area and quiet zone05
The instruction matters after the camera works.
I would also keep the invitation specific. “Scan to see the menu” tells a guest what happens next. It does not promise that payment happens now or that they must place an order before they can browse. The action should match the destination.
That same question continues on the phone. A sign is not a separate branding exercise once it is the entrance to the product. The guest should recognise the venue and find the menu without working out how the physical sign and the digital screen relate.
The January exchange records the prototype filenames; their actual HTML files were not available for this retrospective. The opening request preserves the physical brief; the reading diagram separates the jobs that object has to do.
06
What I would keep from this exchange.
The first attempt was still useful because it exposed a weak brief. I had asked for variation, then recognised that variation was not what made the object complete. My next prompt named the hierarchy and the format that were missing.
That is a part of working with AI I want to keep visible. I can ask for a lot of output quickly. The harder part is knowing when it is answering the wrong question, and being precise enough to redirect it.
The recorded result was a second set of five A5/A6 prototype treatments. There is no verified winner, print test, or scan reliability result in that exchange. The progress was real, but it was progress in the brief and the designs, not proof of a finished table sign system.
07
The first screen begins on the table.
I was asking an AI tool to explore a physical object, but the real interaction began before anyone opened the app. Someone has to notice the card, recognise what it offers, point a camera at it, and trust the destination. A composition that only looks good in a large browser window can fail at the first of those steps. That is why my correction named a hierarchy instead of asking for something more premium. The logo establishes whose table this is. The code gives the action a target. The instruction explains what happens next.
Looking back, I would review those three jobs separately. Cover the instruction and ask whether the purpose remains obvious. Look at the card from a seated distance. Scan a printed sample with ordinary light falling across it. These are the physical checks the saved conversation does not establish. The destination after scanning is another part of the journey. It cannot prove that the January card was legible, printed, or successfully scanned.
The scan is an entrance into an ordering journey. I return to that question in Let me order dinner before you ask me to pay.
Where this chapter landed
What changed.
I rejected the first set, clarified the logo/code/instruction hierarchy and A5/A6 format, and received a second set of five prototypes. The story ends with a better directed exploration. Choosing and validating a production sign remained separate work.
Added while revisiting this story · September 2026
The thinking behind the detail.
Research notes to take these decisions further. These sources were reviewed for this retrospective.
Use the margin as a design constraint
DENSO WAVE specifies the four module quiet zone for standard QR symbols. That requirement supports reserving clear space around the code before composing the rest of the sign. It does not certify any generated artwork or remove the need to test the final printed code.