I wrote eighty lessons. Almost nobody will.

Before this had a name, I had a thing called TinyCTO: about eighty lessons from years of running engineering teams, written by hand over a couple of years, wired up so other people's AI agents could ask it questions. One setting, everything to everyone. It worked well enough that people used it, and well enough that I wanted a second setting, which is how the product I am writing this on came to exist.

Here is the problem with that origin story. I wrote eighty lessons. Almost nobody will.

This is the open question at the middle of the whole thing, and I would rather write it down in public than pretend the onboarding solved it.

Why the blank box does not work

When you ask somebody what do you know, they cannot answer. Not because they do not know things, but because knowledge does not present itself as a list. It arrives when a situation calls for it. Ask a good operations lead to write down what they know and you get five generic sentences, then they close the tab, and both of you feel slightly worse about it.

Ask the same person what went wrong on the last launch and they will talk for twenty minutes, and about four minutes of it is the most useful thing anyone could read about launches.

So the thing to build is not a better box.

What has worked so far

Three things, in order of how well they work.

The best one is other people's questions. When your page cannot answer something, that question queues up for you in the asker's own words, and answering it takes a minute because it is a real question rather than a prompt to reflect. This works because it inverts the problem: you are not recalling what you know, you are reacting to something specific, which humans are enormously better at.

Second is the page interviewing you. One question at a time, in shapes that are easy to finish. Here is a claim, do you agree, in your words. Here is a chapter of your work history, what went wrong in it. What do you tell a junior first. It draws the questions from what you already put down, so they get more specific as you go, and nothing it drafts becomes yours until you say yes to the exact words. This works, and it is still a session you have to decide to start.

Third is your own assistant proposing. It already sees what you write all day. It can notice that you have explained the same thing four times this month and offer it as a sentence for your page. This is the one with the most headroom and the least mileage on it so far, because it depends on what you have connected.

What has not worked

Importing your profile gets you the shape of your career and none of the content. It is worth doing, it takes a minute, and it produces roles and dates. Nobody has ever asked a page a question that a job title answers.

Suggesting topics to write about produces writing that sounds like it was suggested. The difference between a line written because someone asked and a line written to fill a topic is visible from across the room.

And anything that rewards volume is off the table for reasons I have written about separately. A page that pays you for adding things fills up with things added for payment.

The actual question

Can what a person knows be distilled from what they do, rather than authored by hand?

Everything a working person produces in a week is evidence: the decisions in their mail, the shape of their calendar, the thing they explained twice in a document. A system that watched that and proposed three sentences a week, each one either yes or no with one click, would make a good page out of somebody who would never have opened a blank box.

I am fairly sure that is the difference between a tool for people like me and a thing that is generally useful. I am also fairly sure it is a different and larger project than the one I am building, and I do not want to claim it before it exists.

What I will say is that the ordering is not negotiable, whatever gets automated. Something may draft, nothing may assert. The sentence goes on your page because you read those words and said yes to them. The moment a system is allowed to write things about you that you have not read, the collection stops being a record of what you think and becomes a guess with your name on it, and then nothing quoted from it is worth anything.

So: a slow onboarding with a true page at the end, and an open problem I am telling you about rather than papering over.

If you have three things you would tell a junior colleague, that is enough to start. Start there.