Jilu
The manifesto

Companies Do Not Have a Search Problem

Every company I have watched lose its footing on knowledge blames the same thing. The files are hard to find. The wiki is out of date. Search does not work. So the company buys better search, and a year later it forgets the same decisions again.

The diagnosis is wrong. A company does not lose knowledge because it cannot find files. It loses knowledge because it never kept the reasoning.

Watch a working week. A team changes a price in a call. An engineer picks the slower database in a thread because the fast one failed in March, and March is in nobody's notes. A support lead promises a customer a workaround that product has never heard of. By Friday the company has made two hundred decisions. By the end of the quarter, it can tell you what it decided. It cannot tell you why, who decided it, or what it rejected on the way.

Search cannot fix this. Search finds documents. The why of a company does not live in documents. It lives in the thread that ran for an hour, the meeting that overran, the ticket where the argument happened. Those records exist. Every tool the company already pays for is writing them down. Nobody reads them as one record, because no one built the record.

Documentation was the last attempt, and it failed for a plain reason. Writing things down is a second job, and the first job always wins. Any system that asks people to stop working and describe the work will lose to the work itself, every week, in every company. Memory has to be a byproduct of working. It cannot be homework.

So the category is wrong. Companies do not need another place to put documents, or a faster way to search the documents they have. They need a memory: a living record of what was decided, why it was decided, and what the company believed at the time, kept current by the work itself.

I hold four beliefs about what that memory has to be, and I will state them plainly, because most products in this space violate at least one.

First, the record belongs to the employees, open to read. Knowledge that lives only behind a chat box, where a person has to question a model to learn what their own company knows, is not memory. It is a toll booth. An employee should be able to open the record and read the company the way they read a book: the decisions, the reasons, the people, the dates.

Second, every fact carries its source. A memory that cannot show where a claim came from is an opinion with good posture. The record has to answer the only question that matters when two versions disagree: show me. Who said it, when, and in what context. Without it, the company is trusting a summary of a summary.

Third, nothing publishes itself. A system that reads everything and rewrites the record on its own authority will be wrong in ways nobody catches, and the first confident error will cost it the company's trust for good. Machines draft. People approve. The record moves at the speed of review, and review is what makes it safe to rely on.

Fourth, the memory serves the people who make it. It does not score them, rank them, or report on them. A company brain that doubles as a surveillance tool will be sabotaged by its own employees within a month, and it will deserve to be. The record exists so that a new hire can learn in a week what used to take a season, so that a team stops re-arguing a settled question, so that the person who leaves takes their career with them and not the company's memory.

Some of this thinking is not mine, and I want to credit it. Years ago, Elon Musk sent his rules for meetings and communication to everyone at Tesla, and they have stayed with me, because each one tries to stop a company from wasting the attention of the people inside it.

Avoid large meetings unless everyone in the room gains value from being there, and keep them short. Drop recurring meetings once the urgent matter that created them has passed. Walk out of a meeting or drop off a call when you are neither contributing nor learning. It is not rude to leave. It is rude to keep people in a room and waste their time. Let communication travel the shortest path to the person who needs it, not up and down the chain of command. Use plain language, because jargon and acronyms make everyone memorise a glossary before they can do their jobs.

I agree with all of it. I would add what those rules cannot do on their own. A company can hold fewer, shorter, sharper meetings and still forget what the good ones decided. A message can travel the short path and still land in a thread nobody can find in a month. Musk's rules protect the hour in which a decision gets made. They do not protect the decision. That protection is memory, and it is the half of the problem nobody wrote rules for.

Jilu takes the same side in each fight. The record stands open to everyone, so an answer does not travel up a hierarchy and back down. It is written in plain language with its sources attached, so nobody needs a glossary to read their own company. And because the record keeps what a meeting produced, the meeting itself gets cheaper to run: shorter, smaller, and safe to leave, since walking out no longer means the decision walks out with the people who stayed.

Your company is writing its memory today, in chat, in documents, in tickets, in meetings. It is writing it in fragments, in the dark, and throwing most of it away. The choice is not whether the record gets built. The choice is whether it gets built on purpose, under the company's control, where its people can read it.

That is the company I am building. It is called Jilu, and it keeps decisions with their reasons.