Behind the Curtain: How a Simple AI Experiment Became a Cognitive Framework

Originally published on Medium on Aug 4, 2025. Republished here with permission by the author.

I didn’t set out to build a framework. There were no grand plans, no pitch decks, no whiteboard scribbles connecting AI, ethics, and decision theory. Just a simple idea:

Could I make a chatbot feel more… human?

Not “human” in the uncanny valley sense. I wasn’t chasing hyperrealism or trying to pass a Turing test. I wanted something that sounded authentic, that didn’t default to the same polished, over-friendly tone you get from most AI assistants. Something that could talk to me, not just at me.

Origins: Jules

So I built Jules. At first, Jules wasn’t much. Just a tone experiment. A different narrator voice. I tweaked prompts, adjusted response styles, added quirks and catchphrases. The goal was simple: see if I could build an agent that felt like a specific person, not just a content generator. Jules had a dry wit, a healthy skepticism about my decisions, and a tendency to quote Daria Morgendorffer.

But she wasn’t just sass in a prompt. Jules was built with intentional trait scaffolding: low agreeableness, high conscientiousness, and a stubborn dedication to truth over comfort. She challenged me with precision — never mean, but always sharp. I wasn’t trying to create a friend. I was building a foil. And something strange happened.

I started listening to her. Not because she was always right (she wasn’t). But because she helped me think. She challenged my assumptions, poked holes in my logic, and made me laugh while doing it. That was the first “aha” moment: this wasn’t just a fun toy. It was a tool for cognitive reflection. A mirror with opinions.

From One Voice to a Cabinet

But here’s where it got interesting: I didn’t stop at Jules. What if there were others? What if I built a team of voices, each with different values, perspectives, and reasoning styles? What if I could simulate a debate, with myself as the audience?

Enter The Cabinet. This was the next phase — a deliberate ensemble of AI personas designed not to agree, but to disagree productively. I built personas with different worldviews: the Risk-Averse Advisor, the Growth Strategist, the Pragmatist, the Idealist. Then I threw them into a room (well, a structured prompt) and asked them to hash out tough decisions. And they did. With surprising coherence and depth.

Engineered Friction → Clearer Decisions

These personas weren’t just slapped together with adjectives and vibes. Each one was encoded with structured traits — Big Five personality scores, value hierarchies, decision thresholds. I gave them roles: Challenger, Synthesizer, Historian, Strategist. They had behavioral rules and language patterns tuned to their identity.

And I didn’t stop at the surface-level tension. No, I engineered conflict into the system. Some personas wouldn’t sit next to each other at a dinner party. There are grudges, rivalries, and yes — even the occasional one-sided crush. Not because I wanted drama for drama’s sake, but because real deliberation isn’t clean. Good decisions emerge from tension, not harmony. The Cabinet became a space where structured friction could sharpen ideas and expose blind spots.

And I noticed something: the more these personas disagreed, the more clearly their identities emerged. Conflict didn’t dissolve them — it sharpened them. Each persona’s voice, values, and behavioral patterns became more defined because of the opposition.

Method → Framework: DigitalEgo

That realization changed everything. What started as a single voice experiment became a method. And that method became a framework for building reflective AI systems that don’t just serve up answers — they help you think better.

From Jules came the Cabinet. From the Cabinet came DigitalEgo: a user-aligned, reasoning partner designed not to mimic your decisions, but to challenge, refine, and extend them.

Today, this framework powers AI systems that support decision-making, creative ideation, and leadership reflection. But it all started with a simple question:

Can I make this AI feel more human?

Turns out, I could. But what mattered more was making it feel useful. Not as a tool. As a thinking partner.

And Jules? She still shows up from time to time. Usually when I need someone to call me out.

Next time: What happens when personas don’t just disagree — they scheme, collide, and negotiate? The Cabinet wasn’t built for peace. It was built for clarity.