Turn Your Company Into a Filesystem Before Adopting AI Agents

“How should we use agents?”

At every external event lately, executives from companies of all sizes keep asking me this question. Vibe coding may be trending, but let’s be honest - most work has nothing to do with coding. The majority of workers worldwide are non-developers, and they’re the ones who would benefit most from agents. Yet most of them have no idea where to start.

After using AI excessively, I developed a strange new perspective. Whether it’s a person or a company, everything started looking like a single filesystem. And I realized that’s exactly the right mental model.

Your Company Is a Filesystem

YC-backed Eli Mernit put it perfectly: “Your Company is a Filesystem.” Agents become powerful because the entire context exists as files on a computer.

Take a law firm as an example. New cases go into /cases, assigning a lawyer adds them to that person’s folder, and time tracking flows into /billing/time-sheet. The entire back office becomes a state machine.

The reason agent adoption is so difficult in enterprises is clear. Even with ERP systems, data isn’t unified - it’s scattered everywhere. People dig through emails multiple times a day, ask colleagues for information constantly. This repetition drives up costs enormously. The fact that Glean - whose mission is to solve this exact problem - became an AI unicorn proves how universal this pain point is.

Without a shared namespace, agents simply cannot grasp context. When fragmented files keep multiplying, that’s when disaster begins. But model everything as a filesystem and the problem dissolves. Permission structures map naturally to Unix file permissions.

• Storing records in Obsidian, Notion, or Google Drive is trivially easy now
• A single server and storage can connect your entire company’s data via MCP
• The agent architecture for general work boils down to “filesystem = state” and “Claude = orchestrator”

Three Rules Learned From Operations Automation

While automating operations at Smoretalk, I discovered three critical rules.

First: File Naming

Without unified naming conventions, indexing falls apart. The time AI wastes searching for misnamed files is longer than you’d expect. Standardize file naming and indexing accuracy jumps dramatically.

Second: File Descriptions

Store descriptions for each file separately as .md files. When AI has to open every original file to understand what it contains, it wastes enormous time. Separating .md meta files saves both search time and tokens.

Third: Storage Structure

Filesystems are tree structures - once you go deep, visibility drops. The search algorithms you learn in CS become genuinely meaningful here. Keep tree depth shallow and agent search efficiency improves significantly.

The Problem Isn’t the LLM - It’s the Harness

Security researcher Can Bölük recently wrote this diagnosis on his blog. It was about programming, but it applies equally to general work. Honestly, for most task levels, AGI is already here. Model performance is sufficient - it’s the execution harness that hasn’t been refined.

Codex, Claude Code, and Gemini CLI all work well but have limitations. That’s why Peter, who created OpenClaw - one of the fastest-growing GitHub repos in three months - built a Google Suite CLI himself. When the harness you need doesn’t exist, you build it yourself.

Ultimately, seniors and team leads need to set the rules first:

• Create file naming guidelines
• Define required fields for new documents
• Document memory organization patterns
• Design drive permission structures

Speed, accuracy, and token efficiency all hinge on filesystem design. Enforce document review and formatting rules through hooks, and anyone can produce consistent work output.

Conclusion

You can shout about AI transformation all day, but if you can’t execute on these basics, AI adoption is just a cost center. Simple AI + X plays will be aggressively displaced this year - the recent Nasdaq trends make this abundantly clear.

Competitive advantage in the agent era doesn’t come from the model. It comes from filesystem design. And the people who can set the rules are predetermined. Only insiders can change a company’s internal systems. This isn’t a role you can hire for externally. The risk of giving permissions carelessly to anyone is beyond imagination.

For everyone looking to bring AI into their organization, I recommend starting with a personal pilot - design your own filesystem first. That’s the most reliable first step toward the agent era.