Example scenario, not a client project. A field-service company wants technicians to get repair instructions from AI based on a photo of the fault. Instead of building an app, a coded prototype is built in a week on 30 real service tickets. It turns out the AI recognizes the device well but gets older models wrong — so the design has the technician confirm the model before the instruction.
Is this for you?
- The team has been discussing an idea for weeks, and no one has seen it yet.
- You want to show the idea to the board or investors working, not on slides.
- You don't know whether AI can handle your real data.
- You want to test the process with users before building it.
What I do
- Clickable prototype — in a design tool; fast, good for testing flows and screens.
- Coded prototype — a live app built quickly with the help of AI (so-called vibe-coding); it lets you check how the AI really behaves.
- Workflow prototype with real data — the whole process on a few real cases, including the human steps.
How I choose the medium: I start from the question the prototype has to answer. "Will people understand this flow?" — a clickable prototype is enough. "Can AI handle our documents?" — it takes code and real data.
The AI layer
- A prototype with a real model shows things a mockup can't: response time, errors, unpredictable results.
- I check how the interface copes with an uncertain or wrong AI result, not just a perfect one.
- I test the human checkpoints from Workflow & roles on real cases.
What you get
- A working prototype with a shareable link.
- A description of what the prototype tests — questions, assumptions, limitations.
- Test scenarios — ready to use in Testing.
- Lessons from the build — what turned out harder or simpler than we assumed.
How it runs
- The question — what exactly the prototype has to test.
- Choosing the medium — clickable, code, or a workflow with data.
- Build — in short cycles, with a review every few days.
- Demo and first tests — with the team, then with users.
Roughly from a few days (clickable) to 2–4 weeks (code with data) [to be confirmed].
How we'll know it worked
- The question we started from has an answer.
- The team and decision-makers see the same solution, not each their own version.
- It's clear what to build, what to change and what not to build at all.
What this doesn't cover
A prototype is for learning, not for production. It isn't secured, scalable or maintained — production code is written by your team or a vendor.
Visualizations
Which medium for which question
The question picks the medium.
A prototype at the edge (sample artifact)
Sample artifact: a prototype tests the bad day, not just the demo.