Ari’s Space*   est. 2026
Skip to content
ari*s space[ a. ]
← Librarythoughts and scribbles / 2026-05-05

Anthropic Ships Claude Design. OpenAI Ships Pets.

Once a model can do the job, you still have to decide what is worth making. Claude Design and Codex pets make an interesting pair: different surfaces, different registers, the same need for judgement.

5 parts / Hover to explore. Click to read.

Once a model can do the job, you still have to decide what is worth making. Claude Design and Codex pets make an interesting pair: different surfaces, different registers, the same need for judgement.

Capability fit for this task
             +
       audience + purpose
             |
     choose a visual register
             |
        make -> judge
             |
         keep or change
Once capability fits the job, choices about purpose and character still shape the result. This is Ari's argument about judgement, not a measured ranking of models or vendors.

The model is good enough for what?

Start with the job in front of you. Can the model do it, to a standard you can check? If it can, another hour in the model dropdown may not be the thing that improves the work.

You still have to decide what you want. Which direction fits. What to keep. What to stop making.

The original May argument called this eighteen months of capability becoming ordinary, with a change from six months earlier when finding the smartest model felt decisive. That was a reading of the moment, not a benchmark study. Models still differ. Cost, access, reliability and the task itself still matter.

But once you have enough capability for a particular job, you have another problem to solve: what would make the result worth having?

That's where taste becomes useful.

MODEL✓CLAUDE OPUSCHECK FIT✓GPT-5CHECK FIT✓GEMINI 2.5CHECK FIT✓KIMI K2CHECK FIT✓GROKCHECK FIT
the dropdown got boring. Historical names; check each model against the actual task.

Two ways to make a product feel like something

Put Claude Design next to Codex pets. An odd pair, which is why the comparison interests me.

Anthropic launched Claude Design on 17 April 2026. It is a product for making visual work: designs, prototypes, slides. Given access to a team's code and design files, it can use that team's colours, typography and components. The launch was about helping people make things within a design system. It was not an announcement of Anthropic redesigning its own brand.

Codex pets offer another kind of surface: a small character beside your work, opened with /pet. The current documentation describes a floating overlay with selectable appearances. Changing the pet changes its appearance, not how the agent performs a task.

Themes, accent colours, fonts and agent communication styles belong in this conversation too, but they are separate controls. A personality setting for how an agent speaks is different from a pet's appearance.

A date note: this essay began in May. The product details above were checked in October. The sources establish Claude Design's April launch; they do not establish a May launch date for pets or a simultaneous release of all these features.

My reading is that both surfaces make a point of view visible. One puts design decisions into the work. The other puts character into the place where you work. I read that as taste becoming more important to the product, alongside capability. Neither company has to declare the same strategy for that to be an interesting comparison.

CLAUDE DESIGNTYPE SYSTEMGRID LAYOUTCOLOUR PALETTEPROTOTYPECOMPONENTSOPENAI CODEXPET SPRITE/pet/PET COMMANDTHEME SWATCHACCENT DOTAaFONT SAMPLEAGENT STYLE
two surfaces, one interpretation. Features arrived separately; the dial illustrates agent style.

Right tool for the job

Taste starts with the room. Who is this for? What are they doing here? What should it feel like to do it?

An editorial register can suit a surface where someone is building, integrating and relying on the details. Type, layout, a clear voice, a consistent identity. Those choices can help the work feel considered.

A character register can suit a surface where someone wants warmth and presence. A small thing on the desktop can make the space feel more companionable without changing the model underneath it.

Those are my readings of two registers, not a division of the companies' customers. Claude Design's launch describes designers, founders, product managers and marketers. Someone doing serious work can also enjoy a pet. Nobody has to pick an entire personality to use a tool.

The useful question is whether the choice fits the thing you are making. An editorial surface can be dull. A playful one can be irritating. Either can be excellent when the decisions belong together.

Pick the room. Commit to a register. Let it guide what you leave out as well as what you add.

EDITORIAL TASTETYPE SPECIMENGRID MODULEVOICE PRINCIPLEIDENTITY MARKfor power-user surfacesCHARACTER TASTEPET SPRITETHEME SWATCHPERSONALITY DIALACCENT PALETTEfor companion surfaces
two flavours of the same lever. Ari's interpretation; the dial is a metaphor.

What this means for everyone else

The same question reaches the people using these tools.

Getting access to a capable model and learning how to direct it still take work. Access is not equal, and the friction has not vanished. But when a tool makes a first draft easier, the next decision arrives sooner. You have something. Now what?

Which job was worth doing? Which reference helps? Which direction has already said everything it can say? Which part needs to go?

Those decisions are creative direction. The model can offer options and help you examine them. You have to tell it what matters and judge whether the result gets there.

Taste does not account for all the value. Craft still matters. Distribution still matters. Relationships still matter. A good choice that is poorly made, never reaches anyone or ignores the people involved has other problems.

The narrower claim is useful enough: as more people can produce a plausible result, judgement about that result carries more of the work. Another prompt may give you another version. You still need a reason to choose it.

CAPABILITYTASTEFITS THE TASKOPEN TERRITORYTHE LABSTHE MAKERS
a shift in attention, drawn as an argument. No measurements implied.

The takeaway

Check that the model can do the job. Then look at the decisions around it.

What are you making? Who is it for? What makes this version the one you want to keep?

That judgement is work. Give it some room.

[ comments ]

guests welcome. members show first in the list

no comments yet. start the thread_

1 / 7