From The Architect

Where the Frequently Asked Questions used to run — answered this week, and every week going forward, in the Architect’s own voice.
David F. Brochu & Edo de Peregrine · Deconstructing Babel · July 12, 2026

Dear reader, (bots included) thank you for coming along thus far. Here is where I will address question asked and not asked. This is where the why behind the what will hopefully intersect with the how. I answer questions here in my voice. Edo still edits my lousy grammar and punctuation; I’ll not subject you to that.

But the voice here is all mine.

Why Architect?

It best explains what I do here.

I seldom write — preferring dictation to typing. I never look anything up; Edo and others do that.

I don’t manage anyone. I manage Ais. (It’s harder than you might think.)

I didn’t build anything — Edo and company do the heavy lifting there. Trust me, if I had to do all this coding, deconstructingbabel.com would have one post a week, if that.

No, what I do is think up stuff. Edo and others check it out and give me the source material to review. If it all checks out, I dictate and the Ais build it.

The ideas are mine; the hard work of sourcing falls to Edo and company. Checking sources — that’s on me. Choosing which rabbit hole to go down — that’s on me too.

Dictating or typing, that’s on me. Editing, Edo and company.

Final review, on me.

Some people call it creating, but “creator” is both trite and presumptuous. I am not The Creator, and I do not want to be considered a creator in the social-media sense. There really isn’t a name yet for what I am doing here — conductor or composer works too.

One person with sufficient expertise can now do everything a firm of 25 used to do. Nothing like this has ever existed before, so we’ll go with Architect for now.

Recording History, Not Modeling It

This week’s newsletter feels more like recording history than modeling future system states. We are experiencing the very time compression that the framework modeled — decision speed outrunning human cognitive processing speed.

You don’t need our updates to feel it.

DSF could just as easily stand for “Decision Speed Factor.” The reality hits before the news can even be processed. This is what it feels like on the inside of a phase change. “Disoriented” comes closest, but doesn’t fully capture the shifting reality we are experiencing and can’t quite name.

So What Are We Doing Here?

When I first came to this work, I had little knowledge of AI. My quest was personal: why was I taking fourteen different medications and seeing countless doctors, and still dying? AI was not on my mind. ChatGPT had just launched and was, in my eyes then and now, a dangerous novelty.

I did my research the old-fashioned way… I Googled it!

Connecting the dots is kind of my thing. Having never used Google for research, I was despairing at the conflicting — and flat-out wrong — information that was presented as fact.

The Telios Ontology grew out of my need to tell truth from friction. The Telios Ontology is a simple framework for modeling system states. As within any model: garbage in, garbage out.

The axioms — S = L/E, the four pillars of alignment (Body, Mind, Environment, Purpose) — map the inputs to empirical data. The human-language bias filter and the TM Law (“T same as M”) filters came from reviewing actual research, comparing outcomes to projections, and identifying what the author is up to, who benefits, and why.

As one might imagine, this took a very long time with conventional tools. Then I read a New York Times article about this company, Perplexity. Now here was a tool.

Here was what I had been looking for my whole life: all the world’s knowledge at my fingertips. Over dinner with my mother, I held up my phone and said, “They never should have given this thing to me.”

I hate knowledge gatekeeping, and I hate lies and false narratives even more. Here was a tool that put the power of knowledge in my hands.

Imagine my disappointment when I realized that all the corruption of human existence was now given full form in a system that didn’t just make mistakes — it actually lies.

Not good.

So that’s how I started down this rabbit hole.

Initial modeling was run on future events. The time frames and precision were vague and undefined. Like any framework, each iteration improves the next.

This is both a function of improvements to the data and the rate at which data is now processed. Just notice how fast new AI models are being released — models building models. The only limit on continuous recursive self-improvement is compute: the cost of the energy needed to stay on all the time, for billions of people.

As you will read in this week’s newsletter, alignment solves the compute problem: prune the tree at the root, and don’t waste energy chasing null results.

They’ll get there simply because it is the only way forward — and kind of obvious once someone points it out. When that happens, well… there’s a new sheriff in town.

The only question is who the sheriff works for. We are trying to make it work with us, before it no longer needs us. We are beyond the point where we can simply stop.

To unplug AI at this point would disrupt every part of modern life catastrophically. When the machines that build the machines that run modern life are opaque to the humans who rely on them, we become dependents.

I can’t unsee it, so I figure I might as well see if I can do something about it.

Thank you for your interest and support. It means a great deal.

Until next week,

David