-
Interface Beats Model — the throughput-vs-reading wall
AI token generation already outpaces humanity's reading capacity, and the gap compounds ~25%/month. A data check on the Winking claim that the interface — not the model — is the bottleneck. The cross-over is real; the proof that interface beats model still needs A/B evidence.
-
Superpower or Superwaste: What AI lacks is a mission
We built the most powerful reasoning tool in history and aimed it at better marketing funnels. A look at the deployments quietly bending carbon curves — steel, cement, shipping, grids — and an argument for pointing the technology at problems that matter.
-
The Case for Less Software: Why Token Counts Are the New Vanity Metric
In 1982 Bill Atkinson reported '-2000' lines of code and management stopped asking him to fill in the form. Tokens are the new lines-of-code. Why the benchmark for AI tools will shift from how much they produce to how precisely they think.
-
The Doppler Trap: Why Waiting for Better Tech Backfires in Industry
Waiting for better technology is individually rational — until a whole industry waits at once and the gap compounds. The durable bet is not picking the winning platform; it is building an organisation that can absorb and deploy change faster than the field.
-
The Intelligence Grind: Why Tech Scouting Runs on Chrome Tabs and Luck
Technology intelligence fails not through missing information but through missing structure — Chrome tabs, shared drives, and knowledge that walks out the door when someone leaves. The case for systems that are deterministic in method and Bayesian in output.
-
The Knowledge Problem: Why AGI Won't Win and Oracles Will
The dream of one general intelligence that knows everything is a Hayekian mistake. Why the AI that wins will not be the most general model but thousands of private oracles — shaped by proprietary data and institutional knowledge no general model can hold.
-
The Selfie Problem: Vibe Coding Wins Even if Experts Hate It
Vibe coding produces a lot of bad software — and so did the camera-phone, desktop publishing, and the Kodak Brownie before it. Why democratised creation wins anyway, what it still leaves to domain expertise, and where the compute divide limits the story.
-
Winking at a Touchscreen: The Interface Problem in Agentic AI
Agentic AI hands you a finished output and removes the iteration where thinking actually sharpens. The problem is not the model's capability — it is the interface. A case for collaboration over procurement, and keeping humans genuinely in the loop.
-
Data Stops at Capital: Mapping the Climate Investment Evidence Gap
Hard-to-abate sectors produce 13.5% of global emissions but attract under 7.4% of clean investment — here's what the data can and cannot prove.
-
Dig Up Stupid: Investment & Policy in Hard to Abate Sectors
The cleanest climate investments tell the best stories, but the math of impact lives in steel, cement, shipping... the "dirty industries". Facing that gap is an important step to actually reversing climate change.