Design for systems that are not deterministic.
Designing an AI product is a different discipline from designing software, because the thing you are wrapping an interface around is uncertain, slow, and occasionally wrong. We design for that — and we build the brand and production systems around it.
Why AI product design is its own discipline
Conventional product design assumes a deterministic system underneath: the same action produces the same result, operations complete quickly, and the interface's job is to make capability discoverable. None of those assumptions hold for an AI feature.
The output varies between identical requests. It arrives over seconds rather than milliseconds. It is sometimes confidently wrong. And the user has no mental model for what the system can and cannot do, because the input is natural language and natural language does not advertise its own boundaries.
Design that ignores this produces the familiar failure: a beautiful text box that users abandon after three attempts, because nothing in the interface taught them what to ask for or told them when to distrust the answer.
What we design for
Designing for uncertainty
Communicating confidence without pretending to precision. Showing sources so a claim can be checked. Making it obvious which parts of an output are generated and which are retrieved. Trust is a design problem before it is a model problem.
Latency as a material
Multi-second operations need a different interaction grammar than instant ones. Streaming, progressive disclosure, optimistic states, and giving the user something useful to do with the wait rather than a spinner to resent.
Teaching the boundary
An empty prompt box is the least helpful interface ever shipped. Affordances, examples, structured entry points, and constraints that teach capability by demonstrating it rather than documenting it.
Error and recovery
The system will be wrong. The design question is whether the user can tell, how much that costs them, and how cheaply they can correct it. Graceful wrongness is a feature you have to design on purpose.
Human-in-the-loop interfaces
Review, approval, and correction surfaces where the human is genuinely faster because of the system rather than merely responsible for it. Most of the value in enterprise AI lives here, and most of it is designed badly.
Feedback that feeds evaluation
Interface signals designed so they produce usable evaluation data rather than a stream of unattributable thumbs-down. Design and measurement are the same problem approached from two sides.
Brand, media, and the systems behind them
Creative work is a core competency here, not a service bolted onto an engineering practice. Brand systems, identity, graphics, photography, publishing, campaign work, social, voice, video, animation, new media — we do this work and we take it seriously as craft.
The advantage is not simply that we can make creative output. It is that we can also engineer the systems, workflows, intelligence, and production infrastructure behind it: content pipelines that hold quality at volume, asset systems that stay coherent as they scale, and production tooling built for the specific way your team actually works.
And it runs the other way. Creative judgment is what makes technology that people can understand, use, and care about. A firm that can only build produces systems that are correct and unusable.
Not an agency that happens to use AI. Not an AI consultancy that happens to be creative.
Questions we get asked
- ?
What makes designing an AI product different?
The system underneath is non-deterministic, slow relative to conventional software, and occasionally wrong, and users have no reliable mental model of its limits. Design has to communicate uncertainty, make latency tolerable, teach capability through affordances, and make errors cheap to detect and correct — none of which conventional product design patterns handle well.
- ?
Do you do brand and identity work, or only product design?
Both. Brand systems, identity, campaign and publishing work, video, and voice are core capability alongside product and interface design. We also build the production infrastructure behind creative output, which is the part most studios cannot.
- ?
Can you design an AI feature without building it?
Yes, and it is a common engagement. We can also build it. The advantage of the same firm doing both is that the design is constrained by what is actually feasible, and the engineering is shaped by what will actually be usable.
- ?
How does design connect to evaluation?
Closely. What an interface asks users to do determines what feedback it can collect, and that feedback is a primary source of evaluation data. Designing the feedback surface and designing the eval suite are the same problem from two directions, which is why we prefer to do them together.