If AI simply helps us stop sooner at the first workable suggestion, we have gained very little. We have mainly brought forward the moment when we settle.
We want more from this technology. AI helps us take different perspectives, try ideas and examine things we previously struggled to make time for. That excites us. Within a comparable timeframe, we can think further and design more carefully. Sometimes we finish sooner too. Our measure remains how good the result is.
After our article about rights in AI-generated design, we want to explain how we work ourselves. Anyone asking questions about authorship and responsibility should be equally open about their own practice.
The first suggestion has competition here
A convincing concept can bring a discussion to an end surprisingly quickly. It looks good, the flow sounds plausible and everyone can picture it. That is precisely when we find another question useful: What other solution could work?
During concept development, we use AI to put alternatives on the table. We ask for counterarguments, consider flows from different roles and challenge assumptions. Perhaps a new user needs more guidance. Perhaps an experienced user can skip several steps. Perhaps the feature under discussion is unnecessary for the actual task.
The value lies in the discussion that follows. Our designers debate the directions, reject suggestions and develop the promising ideas. AI is welcome to challenge us. A tool that merely endorses our first idea at length offers little help.
We treat a simulated user perspective as a hypothesis. We establish whether it holds through real people and observed tests. Research needs the same attention to its foundations: we trace a claim back to its source before building a decision on it.
We are not sentimental about copy and paste
Transferring values, comparing similar states, preparing test data: these tasks belong to the work, demand attention and invite mistakes through repetition. We have no attachment to spending longer doing them by hand.
AI and automation take on some of that work. We choose tools pragmatically. A script is often the better answer for a fixed rule. AI can be useful for unstructured material, potential contradictions or additional test cases.
We need the attention elsewhere: for an unclear flow, a difficult design decision or an edge case. When a change is applied a hundred times, we check the rule and its result. Repetition does not make a flawed approach any more correct.
Try it early and give the discussion something tangible
Describing an idea and using it are different experiences. A flow can sound perfectly sensible in conversation. Clicking through it reveals how much guidance a step requires or how awkward a correction becomes.
With AI assistance, we can build demos and prototypes early. Consider a multi-step data-entry process: a compact interface and a guided flow can be tried side by side. Long entries, missing information and error messages reveal where each approach holds up. Synthetic data is sufficient to start.
This changes what we can discuss within the team and with clients. A testable model makes decisions more concrete than a description does. Rejecting a direction at this stage is a good result when it saves us from an expensive detour.
Before testing, we define what we want to learn. More test cases alone do not move us forward. We want to know whether someone finds the next step, can correct an entry or understands a state. Automated checks complement manual reviews and tests with people.
The prototype has served its purpose when it answers that question. For production, we then deliberately assess what to retain and what to rebuild. Architecture, permissions, security, performance and operations require careful work of their own.
We create the final design ourselves
In design, we want to make the decisions that make a piece of work ours. Structure, typography, colours, forms and interactions take shape in the design team. This is where the final ideas are discussed, developed and refined.
We care about how those decisions work together. An interface should be beautiful, understandable and convincing in its less spectacular states too. That includes empty views, error messages and the moment when someone does not know how to proceed. These details deserve the same attention as the first screen.
Our designers create the final custom design assets for our clients. The editable source files are retained. AI-generated material can support internal exploration; such a draft is not silently handed over as our final custom design.
This is exactly where we want to invest some of the time gained: in refinement. It would be a shame to stop earlier at the point where a good direction becomes a strong design.
“The AI did it that way” is not an explanation
Anyone accepting a result must be able to assess and explain it. That applies to a design suggestion, a research finding and a code change alike. Responsibility stays with the person approving the work.
Our review therefore follows the task. For a concept, we examine logic and relevance. For design, we look at consistency, interaction and usability. Our developers take responsibility for architecture, code, dependencies, licences and security. That includes work AI helped write or review.
A second pass through the same model is not an independent check. We need to assess the result against sources, requirements, tests and our professional judgement. If something is wrong, we must be able to correct it. Blaming the AI is as unhelpful to a client as it is to us.
What belongs in data handling and handover
We share confidential information only within a suitable setup approved for the project. Where possible, we use synthetic data for demos. Many specialist questions can be described with enough context without sharing personal information. Service terms, access and storage belong in the choice of tool.
Handover depends on our own design work, available source files and clarified rights. For our custom design work, we secure rights within the team so that we can contractually grant you the agreed usage rights. The included source files and uses are defined in the project agreement.
We identify licensed fonts, libraries and other third-party components along with their relevant conditions. The previous Insight examines authorship and rights in depth. Here, the practical consequence matters: you should know what you receive and how you can continue working with it.
This is what we look forward to
AI opens up possibilities we want to use. Try more. Follow an alternative concept. Test an interaction earlier. Hand over repetitive tasks and focus our attention where experience and design judgement matter.
A prompt can set a lot in motion. We still decide when the work is good enough.
