I Went Quiet in April 2023. I Was Building Governed AI Architecture.
- David Harvey
- 3 days ago
- 3 min read
Why I’m ready to talk about what, until then, had only existed in my white papers.
The specification traces back to the 2010–2011 New Zealand earthquakes. By 2014, I’d even designed the physical housing: an earthquake-proof intelligence table.
By 2018, I knew as much about what I didn’t want as what I did. By 2020, those lessons had become a 100-point engineering specification.
In April 2023, I deliberately stepped away from public AI conversations. I’d reached the point where building mattered more than debating. Around that time, Madelyn — a director from a U.S. health insurer — asked the question that ultimately shaped everything that followed:
“Show me the theory working. Why should anyone move beyond Net Promoter Score to the Comfort Index?”
I realised I didn’t yet have an answer I could prove.
That became the standard. Not whether the ideas sounded good, but whether they could survive real-world evidence.
What followed wasn’t three years spent searching for a better idea. It was three years discovering what actually makes AI trustworthy. Today’s AI systems rarely fail because they lack capability. They fail quietly at trust, confidence and evidence. Those failures are often only discovered after decisions have already been made.
A governed AI architecture journey
That changed everything I was building. Instead of another AI application, I started building a governed decision engine — a governed AI architecture where evidence, provenance and governance were front and centre from the beginning. Ask Omega, the Comfort Index and the surrounding intelligence ecosystem grow from that same foundation.
Today, that work is delivered through Ask Omega, the first public pane of the engine, and with an API framework that allows it to integrate with real-world systems. Behind it sits the same architecture we’ve refined through independent reviews, repeated redesigns and continuous testing. Early versions didn’t survive untouched, and that was the point. The goal was never to prove an idea. It was to discover whether it could withstand evidence.
That process continues today. Our recent beta connects the engine to live external systems and passes 1,400+ end-to-end tests on every build. Independent reviews challenged almost every assumption, and we routinely assess every release against our published engineering standard. Recent builds have consistently exceeded 9/10. Those are internal assessments—but every score is backed by repeatable tests, engineering receipts and evidence.
There was one practical obstacle left. The design was waiting for hardware capable of running it privately, locally and without depending on a cloud connection. For years that hardware simply didn’t exist. Apple’s M4 generation is the first platform that finally makes that vision realistic and globally supported.
That’s why this isn’t just another app. During a disaster, it can become a resilient communications hub. At home, it becomes a personal intelligence appliance. In a clinic, it supports governed decision-making. For an extended family, it becomes shared institutional memory.
The same engine adapts to each environment without changing its principles.
The earthquakes were never the market opportunity. They were the forcing function that demanded a different kind of AI—one designed to remain useful when confidence, infrastructure and certainty are under pressure.
That’s what I’ve been quietly building. The difference today is simple: the technology has finally caught up with the architecture.
If that way of thinking resonates with you, I’m glad you’re here.
If you’d like to explore further, the website has been extensively updated and tells the story in much greater detail. Here's the grounding article.
— David