The Story of Keyrk_AI: Part 3
In my last post, I shared how voice-based AI helped me (and others) think more clearly and articulate what we know, especially the kind of expertise that usually stays locked in someone’s head. That process opened the door to something much bigger.
The next “aha” moment came from an unexpected place: a phone call with my old college roommate, David Larson.
We were catching up on life, work, and what’s next for guys like us – seasoned veterans with decades of experience but not quite ready to ride off into the sunset. We talked about how a lot of folks in our position turn to hourly consulting. It’s familiar. It’s flexible. And it’s honest work.
But as we talked more, I found myself explaining why I didn’t want to go that route. It wasn’t just the paperwork and scheduling, it was the ceiling.
There are only so many hours in a day. And no matter how much experience you bring to the table, there’s a hard cap on what people are willing to pay per hour. It’s a model that trades time for money, and at some point, it stops scaling. There’s no way to get out of the “One hour of my time = One hour of my rate = One unit of value”. It’s a straight-line graph.
That conversation stuck with me. I kept thinking:
What if there were a way to multiply the value of what one person knows, without multiplying their hours?
That question sent me down another rabbit hole.
If AI could help me organize ideas, generate content, and surface insights, what else could it do?
• Could it help analyze complex inputs?
• Could it help me build lightweight prototypes or agents that performed tasks on their own?
• Could I use it to test ideas, run simulations, or model outcomes?
So I started experimenting. I explored agentic frameworks, where I assigned specialized role to a set of AI “workers”. I began breaking down complex knowledge into reusable, composable building blocks. And at every step, I asked:
How could AI accelerate or augment this part of the process without losing human judgment where it counts?
And here’s the thing: it worked.
But only when I stayed in the loop.
That became a key lesson. With generative AI and autonomous agents, outputs aren’t always predictable. The same prompt might yield different results depending on context, temperature, or sequence. So building effective systems wasn’t just about throwing LLMs at a problem, it was about designing the right structure:
• Clear roles for the AI
• Well-formed prompts
• Guardrails and checkpoints where humans step in to review, refine, and redirect
That was the real art: not just what the AI could do, but how it fit into a system that combined scale with trust.
While I was building these systems, I also went deeper into how knowledge should be stored and retrieved. I learned about ontologies (great for defining concepts and relationships), graph databases (ideal for connecting ideas), and vector databases (perfect for capturing similarity and context). I realized that no single format was enough on its own.
So I built a hybrid RAG (Retrieval-Augmented Generation) approach that combined all three:
• Ontologies for meaning
• Graphs for structure
• Vectors for relevance
Now, when I asked a question of the system, it’s responses were conceptually aware. It understood the relationships, the intent, the nuances. And it could respond more like a person who actually “gets it.”
At this point, the system wasn’t just a collection of tools. It was a pipeline that could:
• Capture knowledge through voice
• Structure it into rich, connected formats
• And apply it through chatbots, documents, prototypes, or agents to generate leverage on that initial knowledge capture.
It was the beginning of something scalable. Not just for me, but for others.
Not just to save time, but to turn knowledge into leverage.
Leverage would be the piece that would free me (and my clients) from the “One Hour=One Unit of Value” cage. I could embed leverage into my system so that I could deliver value that was far greater than the hours I was putting into it.
One hour of my time could create value that was multiples (or even exponentials) of my normal hourly billing rate.
More importantly, my clients could leverage the knowledge of their experts to generate business value far greater than the time of those experts. By selecting the right AI components and applying the knowledge of their experts in the right way, they could unlock an exponential value graph. I’ll be publishing a paper that goes much deeper into this topic, called “Scaling Expertise: A Playbook for Hybrid Workforces” when I officially launch the company. Keep an eye out for it.
In the next (and final) post in this series, I’ll share how all these ideas came together into a platform, a methodology, and ultimately, a business.