The Filing Cabinet That Waits

Right now you can open a chat window and talk to something that will reason through a contract, catch the flaw in your pricing model, draft a board update in your voice, and walk you through a hard decision better than most of the people you would have paid to weigh in. Then you close the window, and nothing has happened. The contract is still unsigned. The pricing model still lives in a spreadsheet nobody updated. The board update is a block of text you now have to copy, paste, format, and send yourself.

That is the strange thing about this moment. The smartest tool most people have ever had access to is also, for practical purposes, unable to do anything. It can tell you what to do with remarkable clarity. It just cannot reach out and do it. A genius sitting on its hands.

For a while I thought that gap was a footnote, a temporary awkwardness on the way to something better. It is not a footnote. It is the whole story. And once you see what closes it, you stop looking at AI as a smarter chatbot and start seeing the first thing in thirty years that can actually run the parts of a business that software only ever pretended to.

Let me walk through how I got there.

A brain with no hands

Start with why the chatbot feels like a toy. It is not a lack of intelligence. The models crossed the "smart enough to be useful" line a while ago. What they lacked was reach.

Here is the part most people skip past. These systems are not built to sit and talk. The research that goes into training them is blunt about it: modern models are trained and rewarded to use tools, to reach into an environment and act on it. Researchers describe tool use as the way a model recognizes its own limits and pushes past them, calling a function, running a query, taking an action, checking the result, and going again. Tool use is not a feature bolted onto intelligence. For a modern model it is close to the point.

Which means the difference between a glorified chatbot and something genuinely useful is not a better model. It is whether you let the model touch anything. Give the same intelligence a set of tools and an environment it is allowed to act in, and it stops answering questions and starts finishing tasks. The conversation becomes work. The genius gets its hands back.

But a worker with hands and no idea how your business runs is still a temp who walked in this morning. Tools alone are not the unlock. You have to point them at the thing that actually breaks in a company. And the thing that breaks is almost never intelligence.

What actually breaks

Think about the last time something went really wrong at your company. The kind of thing that cost you a client or a month. I would bet it was not because someone was not smart enough. It was because a proposal never went out. An invoice got keyed into one system and not the other. An approval sat in an inbox for three days while everyone assumed someone else had it. A status nobody updated, so half the team kept building on a decision that had already changed.

None of those are intelligence failures. They are coordination failures. The small, boring gaps between people and between systems, where work quietly falls through.

And they are not rare. A Harvard Business Review study of workers at large companies found the average person toggles between apps and windows around 1,200 times a day, losing close to four hours a week just reorienting after each switch. Asana's research on how time actually gets spent puts roughly 60 percent of the average knowledge worker's day on what it calls work about work: chasing status, hunting for a document, re-explaining a decision, moving information from one tool to the next. Not the job. The overhead around the job. In the same research, 88 percent of people said time-sensitive work had slipped through the cracks because of it.

Sit with that. The majority of a company's paid hours go not to the skilled work you hired for, but to the connective tissue holding the work together. That is the real tax on a business. It never shows up as a line item. It shows up as things being slightly late, slightly dropped, slightly off, forever.

The filing cabinet that waits

We were supposed to have fixed this. That is what software was for.

Over the last few decades every function in a company got its own system. A CRM for sales, software for accounting, a tool for projects, another for files, another for reporting, another for the messages about all of it. Each one is genuinely good at its job. Each one is a better, faster, searchable version of a filing cabinet.

But here is the quiet truth about a filing cabinet, no matter how good it is. It waits. It holds exactly what you put into it and does nothing until a person opens the drawer, takes something out, carries it to the next cabinet, and remembers to. Software digitized the cabinets. It never staffed the space between them. And the space between them is exactly where the work falls through.

So companies did the logical thing and bought more cabinets. Then they bought tools to connect the cabinets. And when AI showed up, the first instinct was to bolt a chatbot onto one of the cabinets. None of it closes the gap, because the gap was never inside a tool. It lived in the handoffs between them, the part that still waited on a human to notice, decide, and move. You cannot fix a coordination problem by adding one more thing to coordinate.

The cabinet that opens itself

Now hold both halves in your head at once. On one side, a kind of intelligence that is built to reach into an environment and act, that gets sharper the more tools you hand it. On the other, a business bleeding most of its hours into the coordination between systems that only ever knew how to wait.

Put them together and you get something we have not really had before. Not AI bolted onto one app. An intelligence that lives in the space between the tools you already run.

Picture it plainly, because it is not science fiction, it is just the obvious next step. Your CRM stays. Your accounting stays. Your files, your projects, your reporting, all of it stays exactly where it is. Nothing gets ripped out. But the dead air between them comes alive. When a deal closes in the CRM, the record updates in accounting without anyone re-keying it. The approval that has sat for three days gets surfaced, with a nudge, to the person sitting on it. The proposal that is overdue gets drafted and set in front of you to send. The status that changed propagates before anyone builds on the old one. And when the system is not sure, it does the one thing software never could. It asks.

That is the shift. Not a smarter place to store things. Something that operates the things you already store. A reasonably intelligent system with the reach to run the software, the context to understand your business, and the judgment to move work forward and flag you when it hits the edge of what it should decide alone.

The 60 percent starts to give way. Not because people got replaced, but because the connective tissue finally has something running it other than human memory and good intentions.

So what else can we do

Which brings me back to the cabinet.

For thirty years it waited. That was the deal. You bought better and better places to put things, and you supplied all the motion yourself. Every company you have worked in has been, underneath the software, a set of very good cabinets and a lot of people spending their day walking things between them.

The moment the cabinet can open its own drawers, the question changes. It stops being "what software should we buy" and becomes "what should still require a human at all." Not in a replace-everyone way. The opposite. When the coordination runs itself, the hours it was eating come back, and they come back to the people who were spending most of their day being a router for their own work.

So the real question at the end of this is not whether AI is smart enough. It has been for a while. It is what you do with a business that finally runs its own middle. What you build when nothing falls through the cracks. What you chase when the follow-up chases itself. Where you point a team that just got the better half of its week back.

That is the part I find genuinely exciting, and it is the bet I am making with Navon. The intelligence is here. The tools are here. The only thing left to build is the layer where they finally meet the work.

The cabinet is done waiting.

Sources

1. Workers toggle between apps ~1,200 times a day, losing ~4 hours a week reorienting (confirmed). Harvard Business Review, Murty, Dadlani and Das, 2022. Study of 137 users across three Fortune 500 companies. Coordination-cost anchor.
2. ~60% of the knowledge worker's day goes to "work about work"; 88% say time-sensitive work slips through the cracks (reported, vendor study). Asana, Anatomy of Work Index (survey of 10,000+ knowledge workers). Asana sells coordination software, so treat the exact figure as an interested estimate; directionally consistent with independent findings.
3. Modern models are trained and rewarded to use tools, and tool use is central to how they extend their own capability (confirmed). Agentic tool-use reward research (ToolRM), 2025 to 2026. Basis for the "reach, not intelligence" point.