Overview
AI × UX is a signal-intelligence product for designers trying to make sense of a fast-moving field. AI and UX content arrives faster than anyone can follow, and recency is easily mistaken for relevance.
Instead of building another feed, I designed a Signal Architecture: a model that turns noise into signals, signals into perspectives, and perspectives into decisions. It runs live and self-updating at aixux.dk – ingesting, deduplicating and re-scoring on its own since launch.
AI × UX wasn’t designed to deliver more information. It was designed to create orientation.
My contribution
Product Direction
Product & Interaction Design
Signal Modelling
Information Architecture
Product Development
The team
1 x Product Designer (solo)
Year
2026
Outcome
Product impact
Shipped a live, self-updating product at aixux.dk – it ingests, deduplicates and re-scores on a schedule, and has run on its own since launch
Built the full pipeline solo: multi-source ingestion → normalization → scoring → personalised signals
Instrumented and rebuilt the slowest interaction – switching the trend period went from a 3.9-second freeze to ~50 ms, measured before and after
Design impact
Replaced contradictory, per-component trend readings with one shared signal model – consistent identity, time windows and comparison across Feed, Trends and Tensions
Turned data decisions (canonical identity, normalization, scoring) into UX decisions that made the interface simpler and more trustworthy
System impact
A discovery loop that surprises its designer – its first run surfaced "AI safety" and "government AI", topics I never specified
Tensions extended the product from momentum to divergence, showing where credible sources disagree without declaring a winner
Signal Design Framework
The interface is only one expression of the product. The real design work was creating the underlying signal model.
Frame
Reframed the job from aggregating articles to producing signals – designers don't lack information, they lack orientation.
→ This turned "another feed" into a signal-intelligence problem.
Structure
Designed the Signal Architecture – Sources → Normalize → Filter → Score → Signals → Feed, Trends, Tensions – and defined what a signal is before building any screen.
→ This made the model, not the screen, the primary design object.
Generate
Used AI to generate the components and wire the data: filters, scoring, trend charts and the multi-source ingestion pipeline.
→ This made a solo, end-to-end product realistic.
Review
AI's individually competent components gave contradictory answers – "reasoning" up in one panel and down in another, 42 mentions vs 9 stories. I defined one shared signal model as the single source of truth. The same discipline continued after launch: design rules were kept or rejected by measuring them in the browser, not by principle.
→ This is where the design work concentrated: coherence, not code.
Ship
Deployed live with scheduled ingestion, deduplication and re-scoring, and treated distribution (Share) as a launch feature in its own right.
→ This made it a running product, not a prototype.
Learn
Replaced the fixed topic list with a discovery loop that expands its own vocabulary – surfacing subjects I never named.
→ This turned a maintained snapshot into a living instrument.
Where AI × UX stops
Gartner's analytics ascendancy model describes four steps a system can climb: what happened, why it matters, what's coming, and what you should do. AI × UX deliberately stops at the third.
01 Descriptive – automated
Continuous ingestion from curated sources, deduplicated into one stream.
02 Diagnostic – automated
Scoring separates signal from noise, and Tensions show where credible sources disagree, without declaring a winner.
03 Predictive – automated
The discovery loop surfaces rising topics before anyone names them.
04 Prescriptive – deliberately open
The product points attention. The conclusion belongs to the designer.
Steps 1 to 3 can show their work. What arrived, how it scored, what is rising: each is inspectable, and a user who disagrees can see exactly where to disagree.
Step 4 is different in kind. Even when a recommendation shows its reasoning, "you should do X" makes a choice on the user's behalf: what matters, which trade-off to accept, who carries the consequence. And a system that hands over conclusions trains its user to stop forming their own.
→ The higher a system climbs, the more trust it borrows from the person using it. Leaving the top step open is what keeps AI × UX an instrument rather than an authority.