A Life Has No Wage

The outside ruler

For as long as we have measured machines, we have measured them against jobs. That was sensible, because a job comes with a ruler attached. Someone pays a wage for it. Multiply the wage by the hours and you have a number anyone can check from the outside, without knowing the worker at all.

A life has no such ruler. Nobody pays you to find childcare, to move your mother closer, to fight the insurance form, or to plan a week of dinners around one person’s diet. There is no wage, so there is no outside price, and the hours tell you almost nothing. Ten minutes spent choosing the right school matter more than ten hours of laundry.

Early this month OpenAI published a paper that runs straight into this wall. It is called LifeVal, and it is the company’s attempt to measure how well an AI handles the errands of a personal life instead of the tasks of a job. What struck me is how much the authors had to give up to do it.

1. The wage had to go.

   With no market price for a life task, they asked a few thousand ordinary people what a perfect digital assistant would be worth to them, errand by errand. In the paper’s words, “time spent is often a poor proxy for value.” What rose to the top was not glamorous. It was running a household and caring for other people.

2. The grader had to be the person.

   Each task was written by someone who had done that errand many times and lived with the results for more than a year. That same person wrote the scoring sheet. The reasoning is that nobody else can know what the right outcome was.

3. Talk had to stop counting.

   The assistant earns credit only for what was completed. The graders check what changed in the accounts and the forms afterward, not how good the reply sounded.

The only ruler for a life is the person living it.

Told is not known

The finding most people will skip sits near the back. The researchers handed the assistant the facts it needed in different ways. Stated plainly beside the request, the facts were used. Tucked into the person’s past conversations, the very same facts did about as much good as if they had been deleted. The assistant searched its history the way you search an old inbox, by keyword, and missed what mattered.

That is the Wednesday problem from the kitchen table. You did tell it. Telling is not the same as it knowing. A fact said once in passing is a fact your AI will not find on the day it counts.

So the bottleneck in personal AI is not intelligence. The models are plenty smart, and they get smarter every few months with no help from you or me. The bottleneck is whether your life is written down where your AI can use it, and whether your AI can reach the places where your life happens.

The same place, from the other side

I did not read this paper and then design a business. In January I opened a lab in Charleston where people sit beside a local expert and work on their own AI. We reached the same conclusions from the other side of the table, one appointment at a time. Here is how they line up.

The paperA life has no wage.

Individuals come first. A retiree, a parent and a student sit at the same table as a company owner. Nobody has to be a business to be worth the hour.

The paperOnly the person knows the right outcome.

You are in the room, never on a video call, and you pay after, once you have seen the work and it is right for you. The person holding the ruler decides when it is done.

The paperCredit only for what got done.

Most people arrive with a chatbot. They leave with an AI that is connected to their email, calendar, files and accounts, and that finishes things. That is the Improve in Fix, Improve, Move, and when a connection breaks, the Fix.

The paperTold is not known.

Skills. A skill is your rules written down once, plainly, where your AI reads them every time: the Wednesday thing, the diet, the way you like a letter to sound. It is the difference between having said it and it being known.

The paperThe ruler belongs to the person.

We sell no AI. You bring the one you already pay for, from whichever company you prefer, and it is still yours when you walk out. When people move from one company to another, the chat history mostly stays behind. This paper suggests why that costs less than people fear. The history was the weakest carrier of who you are. The written rules travel.

This is why I have never been able to treat work metrics as the measure of this technology. A benchmark about jobs asks whether the machine can replace an hour of labor. A benchmark about life asks whether the machine knows you. The second question is harder, and it is the only one that matters to the person at the table.

What I expect

1. Every major lab publishes a life score.

   Within two years the personal benchmark sits beside the work benchmark, and it is the one ordinary people quote.

2. “What does your AI know about you?” replaces “Which AI do you use?”

   The brand matters less than the written-down layer on top of it.

3. People keep their rules the way they keep their documents.

   Skills get backed up, carried between companies and handed down. A family’s rules outlive any one subscription.

4. Chat history gets demoted.

   The labs stop treating a pile of old conversations as memory and start turning it into written rules you can read and correct.

5. Sitting beside someone becomes an ordinary trade.

   Local, in-person help with a personal AI becomes as normal as a locksmith or a tax preparer. The ruler is inside the person, so the help has to sit next to the person.

I spent the first part of my career in enterprise software. I co-founded a company backed by some of the largest manufacturers in the world, and everything we built was measured in work. It was real and it was valuable. But I am a personal AI person now, and this is why. The moment I wait for at the table has no wage attached to it. Someone looks at the finished thing and says, yes, that is right for me. That is the whole measurement.

The claim, and where it breaks

Already true

One of the largest labs has dropped the wage as the measure of a life task, let the person who lived the outcome write the scoring sheet, and found that facts buried in chat history are close to useless.

What has to happen

Other labs publish their own versions, on tasks outsiders can inspect. Written rules become portable between companies. People prove willing to spend an hour writing themselves down.

Where I am probably wrong

This is one company grading its own models on a test the rest of us cannot run. And the weakness I am building on may be temporary. If the next models can read ten years of conversation and find the Wednesday thing on their own, writing it down matters less and the trade I describe gets smaller. I would be wrong about the how, though not about who holds the ruler.

Back at the kitchen table, the list of five places is still on the screen. Nothing is wrong with the machine. It was never told in a way it could keep. Write the Wednesday down once, where it is read every time, and the next list is yours. Nobody can price that from the outside. You will know. If you would like company while you do it, bring your AI and sit down with us.

A life has no wage. It has you.