Product: NewsArc (An AI initiative that is now part of the main SmartNews app)

Role: Head of Design (The only product and brand designer)

Timeline: 3 Months (From idea to App Store launch)

The Tiger Team: SVP of Product, CTO, ML & Engineering Squads

The Challenge:

Building for the US Market

SmartNews is huge in Japan, maintaining a monthly active user count of over 20 million readers. However, the platform struggled to win over American readers. To bridge this gap, I drove the strategy to build a brand-new app from scratch instead of dealing with 10 years of old code. As the only designer on the team, I worked directly with the SVP of Product and the CTO to design a generative AI app tailored specifically for how Americans read the news.

The Translation Layer:

Bringing Lofty Ideas Down to Earth

Leadership brought lofty, blue-sky concepts to the table, but those ideas lacked user resonance and fell short of established market expectations, risking a limited product that would fail to stand out in a crowded sea of news apps. Furthermore, the engineering team was accustomed to a highly structured, legacy waterfall workflow and struggled to translate these abstract executive demands into clear technical tasks.

Having designed products at every stage of growth, I knew how to bring both visual excellence and sharp product sense to the project. I stepped in to shape leadership's massive ambitions into a focused strategy that protected engineering velocity. I served as the critical translation layer, mapping out the concrete research and delightful interaction patterns required to bridge corporate ambition with engineering feasibility, keeping the developers inspired and productively tasked.

The Process:

Designing with Live AI Data to Stop a Broken Handoff

Leadership's firm belief that a design wasn't "real" until it was powered by a live LLM led to a chaotic, global operational model. US squads rushed daily builds across timezones so Japanese engineers could code overnight, just to have a live version ready for morning reviews. This was entirely new to the teams, and whenever a feature wasn't ready or leadership changed their minds, engineering capacity was wasted on throwaway code.

I stepped in to solve this by getting closer to the technology. I partnered with the backend team and gained direct access to the LLM outputs, allowing me to stress-test capabilities and analyze live data before drawing a single interface. This flipped the paradigm: I was now actively designing for the AI itself. By grounding my work in real-time outputs, I gave leadership the accurate, interactive validation they demanded while completely shielding engineering from unstable build cycles.

The Reality:

Replacing "Know It When We See It" with Real Data

Relying on a "know it when we see it" mentality, Leadership didn't believe in traditional research. To bring real structure to our fast 90-day timeline, I set up a quick validation process. I dug into older user research the company already had, looked at common app standards, and used my own product experience to skip long research cycles. Once we had a prototype, I set up daily team testing ("dogfooding") and ran quick checks on our live designs. This gave us the fast, real data we needed to fix errors early and make smart product choices with confidence.

Protecting the Product:

Cutting Out High-Risk Ideas

AI features can be unpredictable, and leadership brought a wide variety of ideas that were not fully proven or useful within the news space. They suggested several experimental features that required serious design evaluation and clear boundaries.

• The AI News Map (Dropped): Leadership envisioned a visual map experience powered by the LLM. I dropped this concept after testing proved it violated core user mental models. Readers to know what’s happening now, not pan and zoom around a video game style interface to discover what is happening in the world.

• Generative AI News Summaries (Dropped): Leadership wanted the AI to write news summaries. I monitored other companies attempting this and saw they were consistently failing due to hallucinations and misjudging the tone of serious, sensitive reporting. More importantly, it ran completely counter to user sentiment around news trust, so I recommended shelving it to protect our brand's reputation.

• AI Chat Forums (Dropped): Leadership believed the news would be more engaging with AI-run discussion boards. I recognized this as an outdated community model that testing showed quickly became toxic and argumentative, so I cut the feature.

The Path Forward:

Guiding the Product to Real Value

While steering the product away from those high-risk ideas, I focused on designing an experience inspired by the potential of AI but grounded in reality.

Research showed American readers were tired of misinformation. They didn't want AI-written text; they wanted high-quality journalism. Instead of using AI to create content, I brought the experience "back to basics" to lean into existing brand loyalty for established publishers.

I designed Arcs & Collections, a system that uses AI to group, contextualize, and sort real articles from trusted sources. To create harmony with hard-hitting news, I focused on the specific topics people naturally enjoy, like sports and lifestyle. By allowing users to create personal "playlists," they could easily follow their favorite publishers and interests, which we then served up through smart, automated curation. This redirected our technical capabilities away from generative risk and toward smart utility, answering the market's demand for editorial integrity.