The Stack

Modern news is broken. Long-form investigative journalism is dying because nobody has time to read it. Meanwhile, short-form hot takes lack the depth needed to actually understand complex issues.

I wanted to answer a question: "Why does Times Radio lose £15M a year — and why might that be genius?"

The research ran to 70 pages across 3 different AI models. Nobody would read that. But the insights were too valuable to compress into a tweet.

1. Triangulate the Research

I used Claude, Gemini, and ChatGPT to research the same question from different angles. Each model surfaced different sources and perspectives. The synthesis revealed patterns none caught alone.

2. Find the Hidden Thesis

Buried in 70 pages of research was one unifying insight: Times Radio isn't a radio station that loses money — it's a content engine that powers The Times' entire editorial strategy. The losses are an investment, not a failure.

3. Design for Depth on Demand

Instead of forcing readers to choose between "too long" and "too superficial," I created three layers:

• Skim — Headlines and key takeaways. 2 minutes.
• Read — Full narrative with context. 15 minutes.
• Audit — Every claim linked to source evidence. For skeptics.

Built in under 3 weeks using AI agents as my construction crew:

• Evidence chips — Every factual claim has a clickable chip linking to the source
• Interactive calculators — Readers can adjust assumptions and see how conclusions change
• Depth toggles — Switch between Skim/Read/Audit without losing your place
• Semantic sectioning — Content grouped by concept, not by source