Hi Friends,

Last month Dustin and I sat down in person for 4 hours and mapped out where every file at Allenix lives in our Google Drive.

It was important for us to get it right. Our email, our Slack, our CRM, our calendar, our code and our website all feed each other, and our agents reads from one and writes into another. We knew if we got the foundation wrong, it will confuse our agents, make mistakes, and light our tokens on fire.

I must say, it has the been the most impactful AI system to date. I see the benefits of it daily and I thought it would be helpful to show our exact file structure, the 4 tests we put it through, and steal our company filing system for your company.

I started my career in investment banking, inside other people’s data rooms. Give me 10 minutes in one and I could tell you whether that company would sell at a premium or a discount. It a was a good first impressions test. The things that drove me crazy (trigger warning):

I still believe how much care a company puts into its data room says alot about how it is run.

So when Dustin and I sat down, I wanted to obsess over ours, especially now that agents read it as often as people do.

Our company data room (filing system) sits at the center of the ecosystem. It is the context layer, and every other tool reads from it and writes back into it. There is also a backup created in Github for added protection.

Let’s walk a real sales opportunity example…

The account research, the deck and the proposal are all built from the same folder, which is why the data room is the heart of this and not a place we put things. Our agents work from an organized library, so they find the right thing at the right time.

We keep hearing “context is king”, but this data room problem is the perfect epitome of why nailing the context is so important for companies today.

Anthropic, who builds the models we use most, put it plainly last September: context is “a finite resource with diminishing marginal returns.” Feed a model more and its recall gets worse. The job is finding the few things that actually matter.

Also, UNSW Business School studied why agent rollouts stall and found the model was rarely the problem. The failures had agents “fed fragmented data, deployed on broken workflows.” Their line, which I would put on a wall: “Capability did not close the coordination gap.”

You cannot hand an agent the few things that matter out of a company data room where nobody ever agreed what goes where.

We wrote these 4 design principles on the white board when defining our folder structure in our Google Drive:

Every folder got a number, so what we open daily sits on top and what only founders should open sits last. We also numbered it thoughtfully to so agents can efficiently find our most active folders on top (great token saving hack). Every client folder got the same 4 subfolders, no exceptions.

Dustin wrote a script that audits the our company drive every evening for anything sitting where it shouldn’t be. It recommends to auto-archive files that may cause confusion. Any issues it finds goes to Slack (example of a daily), I decide the right answer, and that decision updates our overall Knowledge layer. This is the self-healing part which I find really exciting.

If you are a startup and not quite sure how to organize your company’s data room, you can steal ours below. Our parent and sub-parent folders. Just copy/paste the prompt below.

Til next time,

-Ali

P.S. If you would like for our help with building AI systems like this, let us know by responding to this email thread. Happy to help. I read every response.

About Me: I’m Ali Mamujee. I secretly love crime documentaries (does anyone even read this far?). I also run Allenix, where we build AI revenue systems for B2B companies between $5 million and $25 million with complex sales cycles. Praying the Texans don’t go 0-2 this weekend.