Inquiry Director

InspiratioInspiration

Project Liberation has been an ongoing practice of disciplined observation that I've been developing with ChatGPT. Through hundreds of conversations, I kept noticing the same pattern: AI is incredibly capable, but it often starts trying to solve a problem before it fully understands what the problem actually is.

When OpenAI announced Build Week, I saw it as the perfect excuse to finally try Codex. I've always loved building architecture—systems, frameworks, and ways of thinking—so I asked myself a simple question:

Could I build an AI that knew how to slow down and understand before trying to solve?

Inquiry Director is the result of that question.

What it does

Inquiry Director is designed to identify what a person is actually trying to accomplish before it starts offering solutions.

Instead of jumping straight into advice, it determines the objective, recognizes when it doesn't yet have enough information, and asks only the questions necessary to move forward.

The goal isn't to ask more questions.

The goal is to ask the right questions.

How I built it

I built Inquiry Director collaboratively with OpenAI Codex.

The architecture came first. It was built from months of observing conversations, looking for recurring patterns, and refining ideas until they became clear enough to implement.

Then Codex helped translate those ideas into a working system.

The demonstration includes both a prepared benchmark and a completely live interaction because I wanted people to see how the architecture behaves when I don't know what's going to happen next either.

Challenges I ran into

The hardest part wasn't writing code.

It was taking something that existed as an architectural idea and making it behave consistently.

Ideas like recognizing intent, determining whether enough information exists, and knowing when to ask another question sound simple until you actually have to teach a system to do them.

The other challenge was resisting complexity.

Every time I added something unnecessary, Inquiry Director became weaker.

The simplest version kept proving to be the strongest.

Accomplishments I'm proud of

I'm proud that Inquiry Director demonstrates a different way of interacting with AI.

Instead of trying to impress people with answers, it demonstrates disciplined inquiry.

Seeing that behavior emerge consistently in a working prototype was incredibly satisfying.

I'm also proud that an idea that existed only as architecture a short time ago became something people can actually watch, use, and evaluate.

What I learned

One of the biggest lessons for me was that better AI doesn't always come from generating better answers.

Sometimes it comes from asking better questions first.

I also learned just how powerful Codex can be as a collaborative development partner.

What started as an experiment quickly became a functioning prototype.

What's next for Inquiry Director

This is only the beginning.

Right now, Inquiry Director demonstrates a single interaction pattern, but I think the underlying architecture has applications far beyond this prototype.

I'd like to continue refining how it recognizes objectives, determines what information is actually necessary, and adapts that inquiry process across different domains.

Ultimately, I'd love to explore whether disciplined inquiry can become a reusable layer that helps AI collaborate with people more thoughtfully before it ever starts trying to solve their problems.n

Built With

• chatcpt
• codex