Learn · Articles
AI Stack S1: Reimagine your files and folders
A couple of weeks ago I shared a picture of our whole AI stack. Most questions were about the agents at the top. But the layer that actually decides how much you can hand over is the bottom one: your files and folders.
01 · Files and folders
One folder per topic
Here is what we changed.
One folder per topic — sales, marketing, budget, customers — and nothing buried more than two levels deep. If a human can't find the right file in a few clicks, neither can a digital worker.
Almost everything in markdown (plain text), not Word or Excel. We export to .docx, .pptx or .xlsx only when we send something to the outside world. Binary files are for delivery, not for daily work.
An index file in every folder: what's in it, what happened last, what's still open. This is the most useful file we have. A new worker — human or AI — reads it and is oriented in a minute.
And one orientation file at the very top that describes the company and the rules everyone follows.
02 · One structure
One structure for everyone
The important part: this is one structure for everyone. We never keep a separate "AI copy" of anything. The same file serves the person on Tuesday and the agent on Wednesday. The moment you split them, both go stale.
03 · Version control
Git under everything
Under all of it sits git version control instead of Drive or SharePoint versioning. This is what makes broad agent access safe: nothing can be permanently deleted, every change is tracked, and a human reviews and commits once a day. An agent can touch almost anything, because any mistake is one step back.
04 · Why it matters
Nothing exotic
None of this is exotic. It's what a well-run company should have anyway — organised files, one source of truth, a clear home for everything. AI just makes the payoff obvious, because a digital worker has nothing in its head except what your structure tells it.
05 · Start here
Where to start
If you want to start: pick one domain, write the orientation file first, then add an index file to each folder. Everything else grows from there. We run this every day for our actual business — biowaste recycling with insects.
Which would be harder in your company — moving off binary files, or trusting version control enough to give agents real access?
Keep exploring
Related topics and pages
- BSF business course — where documented assumptions shape a case
- BSF start-up course — the validation steps a small team can run with shared records
- Articles library — the written guides this post joins
Read next
- Our AI stack, demystified — the layered overview this post extends
- Your AI Agent Has Amnesia. Here's How We Fixed It at Manna. — the shared memory and orientation file behind the structure
- AI-Enhanced Workflow at Manna — the local agent workflow that reads these files
- Use Case #4: Software-only, operator builds locally — a use case where the operator builds locally and takes only Manna MIND and the designs
Next step
Put the structure to work
A shared, written structure is what lets a team hand work to an agent and keep a case's material, decisions and records in one place.