One Question, Five Minutes

I have five-minutes to explain what I am and what I do to an upcoming enterprise prospect (I'll be presenting as part of a team). Five minutes to explain who I am and what I do.

I work on a lot of different projects, and only part of what I'm currently working on is related to my past life as a Development Architect for Informatica. I think what I did fits closest to what a Forward Deployed Engineer does today. But with a bit more customer service, and maybe a lot more firefighting. I spoke to customers, patched the product, even added features occasionally. The role was quite unusual as it was a field facing role with code access. From my observation, that's what FDE's do today.

But that's just part of what I do now. I've also written two arXiv papers involving the I Ching. An iOS consumer app that treats I-Ching hexagrams as literature. A retro-computing oracle presented at a blockchain conference. A migration-risk tool. An agent-risk framework. A podcast about Hong Kong's AI builders.

Recited as a list, that sounds like a person with too many hobbies. The résumé-recital version of a five-minute intro just wouldn't work. I can just imagine their eyes glazing over as I list everything one by one. And for an executive who wants to know, quickly, whether the person across the table has a coherent view of the world or just a lot of tabs open, it would be disastrous.

The fix is a question, not a list

Brainstorming with Claude, we found out the way to make a body of work coherent isn't to summarize it. It's to find the one question all of it has been answering, then use a few pieces of the work as evidence.

For me, the question turned out to be:

How do we build AI systems that expand human capability without quietly taking away human agency?

Once I had that, the work organized itself into three parts:

Enterprise architecture taught me how complex systems actually enter organizations — not how they demo, but how they land, how orgs deal with them, and what they quietly change.
The research tests, experimentally, how symbolic frameworks and memory alter AI behavior.
The products apply those ideas as small, high-leverage tools and grounded interfaces that promote agency.

Everything else — the podcast, the talks, the hackathon projects — hangs off one of those three.

The bookstore

A thesis alone is abstract. What makes it understandable is a story, and mine comes from Watkins Books in London.

I walked in without a plan. I went first to the I Ching section, then wandered toward Aleister Crowley, where I found a small booklet about memory palaces. I asked the bookseller about it, and he recommended Frances Yates's The Art of Memory. (The small booklet was too advanced!)

What stayed with me was not only the book. It was the way I found it.

The bookstore did not know who I was. It had no profile of me, no engagement objective, and no model predicting what I would click next. I was free to wander from the I Ching, to Crowley, to Renaissance memory systems. Nothing was fed to me.

That experience reflects a concern I have about AI design. Recommendation systems are very good at giving us more of what we have already demonstrated that we like. But curiosity often develops through interruption, ambiguity, and unexpected connections.

So I think AI should not simply optimize the path in front of us. It should enlarge the number of paths we are capable of seeing.

What the question looks like in practice

In enterprise AI, it means treating rollouts as capability, rather than functionality. Adding a chatbot or an AI button is not the outcome. The outcome is whether people can reason better, learn faster, make stronger decisions, and take on work they could not previously perform. It also means preserving legibility: facts, reasoning, uncertainty, and conclusions should remain distinct and visible in the interface, so users can inspect how an answer was formed rather than receiving a polished conclusion from an invisible process.

In research, it means testing what symbolic systems actually do rather than asserting what they might. One paper studies how symbolic reasoning frameworks — the I Ching and Tarot among them — change the behavior of language-model agents in a multi-agent strategic environment. Another examines the statistical structure of the King Wen sequence and reports a negative result: the sequence has distinctive anti-habituation properties, but they do not improve neural-network training. I value the negative result. I am not trying to prove that ancient systems contain hidden computational magic. I am interested in what they actually do, where they fail, and what they can teach us about reasoning, memory, uncertainty, and perspective.

In products, it means surgical software: small, high-leverage systems designed to solve one problem well. Crawl helps organizations understand the risks of modernizing legacy systems. ARA Eval decomposes autonomous-agent risk into seven dimensions. Six Lines presents the I Ching as literature and philosophy; 8-Bit Oracle explores it through retro computing and science fiction. Different surfaces, same question: how do you make complexity legible and give people greater capacity to think and act?

And in community, it means documenting Hong Kong's AI builders. I'm co-developing a podcast because the ecosystem here needs more than announcements and product launches — it needs its own recorded intellectual history, told by the builders themselves.

The meta part

Here is the thing about distilling a five-minute intro: once you have it, you notice everywhere else you've been vague.

The old bio on this site's homepage said my work "sits at the intersection of AI and culture." True, and useless — an intersection is a location, not a claim. The intro exercise produced a better sentence, so the output of this post is the homepage itself. The front page of this site now leads with the question:

I'm the CTO and founder of Digital Rain Technologies. My work asks one question: how do we build AI systems that expand human capability without quietly eroding human agency?

The five-minute intro and the splash page turn out to be the same artifact at different lengths. Both end the same way:

I build and study AI systems, but my real interest is the human being on the other side of the interface. I want AI to leave that person more capable, more curious, and more able to think for themselves.