AI Prototypes
Working prototypes, built to find out what AI-assisted design actually does
- Client
- AI Prototypes
- Sector
- Self-initiated / applied
- Year
- 2025 — 2026
- Role
- Design and build
- Services
- AI-assisted prototypingVibe codingProduct designRapid prototypingNext.js
A running set of working prototypes, built to find out what AI-assisted design changes about the work rather than taking the claims on trust.
- Six prototypes
- Shown here
- Claude Code, Cursor, Bolt
- Built with
- Next.js on Vercel
- Stack
- Ongoing
- Status
01Why build them
Most of what gets said about AI in design is said by people who have not shipped anything with it. I would rather form a view by building, so these exist to answer specific questions: what does the tooling actually shorten, where does it quietly make things worse, and what still needs a designer.
They are working software, not mockups. Several are deployed and in use.
02The stance
The interesting gain is not speed. It is being able to put a real, clickable thing in front of someone early enough that the conversation is about the problem rather than about a static screen.
The risk is the opposite of the promise. Used uncritically, these tools produce plausible work that nobody has actually thought about, and the skill of thinking it through atrophies. I use them deliberately, and I throw a lot away.
The prototypes
Self-initiated
Stakeholder Map
Map stakeholders by dependency, influence and relationship health, flag the ones needing attention, and export a PDF for a 1:1 or coaching session.
Open liveClient work · not shown
Impero
Control triggering workflow
A prototype for the logic that decides when a compliance control fires.
Self-initiated
RoleMuse
A job application assistant: find your fit, shape your pitch, land the role.
What the AI doesPersonalised guidance for each application.
Open liveClient work · not shown
Impero
Task suggestions from a description
Proposing the right compliance tasks from a plain-language description, so the user starts from a draft rather than an empty state.
No screens
Self-initiated
JayBee
Qualitative feedback collection.
What the AI doesGenerates follow-up questions from a response, so a single answer can be probed further without a moderator present.
Outcome
- 01Working, deployed software rather than mockups
- 02A practical read on where AI-assisted tooling helps and where it does not
- 03Prototypes used to test product ideas at Impero before committing to build
- 04A habit of throwing away the plausible-but-unconsidered first pass
Bilhandel.dk
Overhaul for Denmark's leading car marketplace
View case