made by MakeBetter
November 2026
An open letter
Don’t start with AI.Start with the work.
To everyone who is responsible for how the work gets done,
You may be the owner, the operations manager, the plant manager, or simply the person everyone calls when something is stuck. Whatever the title, the job is the same. The work has to get done today, and it ought to get a little easier every year.
Lately that job comes with a new pressure. Everyone tells you AI will change your operation. Suppliers, the board, the news, someone at dinner. Very few of them can tell you where.
This letter is about the where.
Most companies have never seen how they run. Not because anyone was careless. A company is not designed. It accumulates: a column added to a spreadsheet years ago, a phone call that became a habit, a check one person still does because something once went wrong. It works, and it lives in people’s heads.
Margin note: yours included
Ask how an order becomes an invoice and you will get a different answer from each person it passes through. Every one of them is right about their own part. Nobody holds the whole.
For a long time that did not cost much. Software could only follow rules, so you automated the tidy parts and people did the rest. AI is different. It can read, write and listen, which means it can take on the untidy parts: the email, the voice note, the scanned sheet, the number copied from one screen to another. In most companies that is most of the work.
So for the first time, the question “what should we change?” has more answers than you can afford.
Margin note: a good problem. still a problem.
And companies choose the way people choose under pressure. The loudest complaint. The best demo. The thing a competitor announced. Some of those bets pay. Others automate a task that took ten hours a year, while three minutes a pallet, forty pallets a day, goes on quietly costing six hundred. Nobody was foolish. Nobody could see.
You cannot improve what you cannot see.
That sentence is old. What is new is that seeing has become cheap.
Writing a company down used to be a project. Consultants, interviews, workshops, months. It cost so much that you did it once, and it was out of date before it was bound. Most companies sensibly never did it at all.
But the same AI that can read an order email can listen to a person describe their morning. It can ask the two or three questions a good colleague would ask (how many, how long, what goes wrong) and turn the answers into a step on a map, with its time, its person and its system. Five minutes of someone’s day, in their own words.
Margin note: no workshop. no clipboard.
Do that across a company and within a week you have what the workshops never gave you: how the work really runs, counted in hours. And because correcting it takes one sentence, it stays true.
That is what we built. We call it Portal. We built it for ourselves first, because we run a trading desk and wanted to know where our own hours went before we spent anything on changing them.
Margin note: our desk was company number one
Here is what we believe, so that you can hold us to it.
The people who do the work are the only ones who know it. So ask them, and let what they say stay theirs. In Portal a conversation belongs to the person who had it. Everyone else sees what changed on the map, never the chat. People tell the truth about a bad Monday when the bad Monday is not on show.
Count the work, not the people. We measure hours, step by step. There is no ranking of people and no timer on anyone’s screen. The work stays. Who does it changes: by hand, with a tool, or by a tool alone.
A number should say where it came from. Someone said it, someone counted it, a system measured it, or it is a guess. A guess should look like a guess until somebody firms it up.
And AI should end in a decision that a person makes. Not a dashboard. A short list: this change gives back this many hours a year, for this much effort. You choose. Then the same steps are measured again, and you find out whether it was true.
We will not promise you a percentage. We have not seen your company. Neither, in full, have you.
So here is what we suggest. Before the next AI purchase, the next pilot or the next strategy day, spend a month looking. It starts with one call, in which you tell us what the company does. A first draft of your map follows, usually the same day. Then your people correct it, five minutes each, in their own words.
Margin note: one call. then five minutes a person.
At the end of that month you will know where the hours go and what is worth building first. What you do with that is yours to decide.
With respect for the work,
The people who make Portal
at MakeBetter
P.S. If you take one thing from this letter, with or without us: ask one person to describe one task from start to finish, and write down the minutes. You will know something by Friday that the company has never known.