Exploring AI for organizational knowledge
01 — Overview
How can we get a clear picture of who is working on what, and who is actually available?
Companies generate valuable knowledge through everyday work, but it becomes fragmented across email, Slack, GitHub, documents, project tools, and people’s memories. Finding the right information often means knowing where to look or who to ask.
As part of this internal R&D project, I worked with a small team exploring whether AI agents could connect knowledge from across the organization and make it easier to find and use. We were exploring the product and technology at the same time, so building working experiences became part of figuring out what the product could be.
Developed as an internal platform.
The interfaces below are abstractions, not screenshots.
02 — The problem
Unlike a conventional product project, we didn’t begin with a fully defined solution. We were exploring the technology and the product at the same time.
The questions were fundamental. What information is useful? Who needs it? How should an agent surface it? What happens when information from many systems becomes visible together? And what boundaries should exist around that knowledge?
We had to build in order to learn.
03 — My Approach
I treated building as part of the research. Working ideas made abstract questions concrete enough for us to test, challenge, and change.
My focus was the layer where people interacted with the underlying AI systems. I built the frontend foundation and workflows around prompt editing, previews, model aware history, authentication, session handling, and data and state architecture. As the prototype evolved, I could adapt the experience quickly as we learned what worked and what questions we needed to explore next.
Connect the sources
Many systems, one project tree
Knowledge arrives from email, chat, code and documents. The agent assembles it into a project view while preserving the permissions of the systems it came from.
Iterate on the prompt
Every version remembers its model
Prompts are edited constantly, and a result only makes sense with the model that produced it. History stores both, so previous versions can be understood and restored later.
Find the boundaries
The answer depends on who’s asking
Once knowledge from many systems is in one place, who is allowed to see it matters. The same question can produce a different answer depending on who is asking.
Get the answer
Ask who is working on what
The agent summarizes what a person is currently working on, assembled from the systems they actually write in. The summary is generated, so it names its sources and can be wrong out loud rather than quietly.
04 — In hindsight
Building the prototype changed the questions we were asking.
Bringing organizational knowledge together made its potential much easier to see, but it also made questions around access, permissions, and boundaries much more concrete.
The technology has moved extraordinarily quickly since then, but the lesson still holds. Prototypes are valuable not just because they prove something can be built, but because they reveal what you should be thinking about next.
For me, building has always been part of thinking.