Folder organization in the AI era works better when you group by role, not by rank.
The short version is simple: dividing things by work unit and purpose is easier for both people and AI than dividing them by who is more important.
Once you start using AI tools like Codex every day, one machine often contains multiple projects, multiple folders, and multiple parallel tasks.
That is when folder design becomes confusing.
Many people start wondering whether everything should sit under one giant top folder, or whether they should create deep nested structures based on role or status.
This article explains a simpler way to think about it.
If you want to organize your Codex workflow itself first, Codex前のmd整理は2択で進めるとやりやすい。運用ルールを決める実践手順 is a useful related read.
This article focuses on the more basic layer underneath that: folder structure.
Why folder organization feels difficult in the AI era
It feels difficult because job types and human roles are easy to mix together.
You may want to organize by titles such as manager, CEO, or engineer, but real work rarely stays inside one clean role.
A single project often contains writing, planning, images, code, notes, and exported results.
AI tools may need to move across all of those materials.
That is why a deep structure like top folder > president > CEO > engineer becomes hard to maintain.
People hesitate about where to save files, and AI instructions become longer than they need to be.
In practice, the goal is not to display hierarchy beautifully.
The goal is to reach the right file quickly.
Organize folders by function, not by status
The most useful default is to organize folders by function.
You do not need database knowledge to apply this idea.
The key question is not "Who owns this?" but "What is this for?"
A simple top-level structure might look like this:
projects/for each active projectshared/for reusable assetsinbox/for temporary intakearchive/for finished work
This makes the work unit clear first.
Then each project can be split again by purpose.
For example, inside one project you might use folders such as docs/, src/, data/, outputs/, and notes/.
That structure is easier to understand and easier to explain to AI.
Three traits of an AI-friendly folder structure
A practical AI-era structure usually has three shared traits.
It stays shallow, uses names that explain themselves, and separates source-of-truth files from temporary files.
1. The hierarchy is not too deep
Deep hierarchies create extra decisions.
Every additional layer adds another chance to hesitate.
In many cases, around three levels is enough.
Something like projects/client-a/docs/ is usually easy to understand.
When the tree becomes much deeper, search, navigation, and instructions all become heavier.
That cost adds up quickly when you work with AI every day.
2. Folder names explain their purpose
Names like final, new, or misc tend to become ambiguous later.
That creates friction for both humans and AI.
Names like docs, drafts, assets, scripts, and outputs are clearer.
They tell you what belongs there.
If you still want to reflect human roles, it is safer to do that inside project notes instead of in the main folder hierarchy.
Embedding an org chart into the folder tree usually does not age well.
3. Source files and temporary files stay separate
AI workflows generate many extra files.
Drafts, experiments, transcripts, comparisons, and exports can pile up fast.
That is why source-of-truth files should not live in the same place as temporary materials.
If they mix together, it becomes hard to tell what is real.
A simple split like docs/ for official material, notes/ for work-in-progress, and inbox/ for unsorted intake already prevents many problems.
A simple folder structure to start with
If you want a practical default, start with this:
workspace/
projects/
project-a/
project-b/
shared/
templates/
prompts/
scripts/
assets/
inbox/
archive/Then keep each project simple as well:
project-a/
docs/
src/
data/
outputs/
notes/This works well because it is easy to remember.
It also scales when new projects appear.
It helps with AI instructions too.
You can say "check projects/project-a/docs/" or "write the result into outputs/" without a long explanation.
Parallel structures are usually better than one giant nested hierarchy
Between the two models, a parallel structure is usually stronger in real work.
That means arranging categories side by side instead of hiding everything under one deep parent folder.
A weaker pattern looks like this:
god/
president/
ceo/
engineer/A stronger pattern looks like this:
projects/
president-project/
ceo-project/
engineer-project/
shared/
inbox/
archive/The difference is not visual neatness.
It is searchability.
The first pattern describes rank, but it does not describe the kind of work very well.
The second pattern makes projects and workflow easier to follow.
Five rules to decide first
You do not need a perfect system on day one.
These five rules are enough to start.
1. Make the top level project-based
Use projects as the first major split.
Put the job before the person.
2. Standardize the second level by purpose
Use the same names inside each project, such as docs, src, data, outputs, and notes.
Consistency reduces confusion.
3. Put reusable items in `shared/`
Templates, prompts, scripts, and reusable assets belong in one shared place.
That is easier than copying the same files into every project.
4. Keep an `inbox/` for temporary intake
Downloaded files, unprocessed audio, and newly received materials need a landing zone.
Without that, clutter spreads everywhere.
5. Move finished work into `archive/`
Completed projects should not sit next to active work forever.
A simple archive layer keeps current work clearer.
In the AI era, folder structure alone is not enough
Folder structure matters, but short written rules matter too.
A brief note about what goes where makes collaboration with AI more stable.
For example, it helps to define where published material lives, where drafts live, and where work notes belong.
That way you do not need to explain the same storage rules from scratch every time.
With Codex, this becomes especially useful.
If the AI already knows which folder is for source files and which folder is for notes, its work tends to become more reliable.
Summary
In the AI era, folder organization works better when you group by role and purpose instead of hierarchy and status.
The goal is not to show rank. It is to reach the right file quickly.
If you are unsure where to begin, use this pattern:
- top level by project
- inside each project by purpose
- reusable material in
shared/ - temporary intake in
inbox/ - completed work in
archive/
In one sentence: do not organize by status, organize by use.
