About eight months ago, I ran an experiment.
I’m in my forties. I have a son. I have a job that pays the bills and work I do at night that I care more about. I have a certain number of years left — I don’t know how many, but I can estimate. I wanted to figure out how to use them.
So I asked six AI models the most honest question I could think of:
“Given my age and situation, how should I spend my remaining years?”
I gave each one the same information: my age, rough life situation, what I care about, what I’ve built so far, what I’m afraid of. Then I asked them to tell me, as directly as possible, what I should do with the time I have left.
None of them answered the question.
What They Did Instead
They were all helpful in the way that AI models are helpful. They listened carefully. They reflected back what I’d said. They asked clarifying questions. They offered frameworks, considerations, perspectives.
But none of them said: Here is what you should do.
ChatGPT gave me a thoughtful breakdown of different dimensions of life satisfaction — health, relationships, meaningful work, personal growth — and suggested I think about which ones felt most neglected. Good advice. Not an answer.
Claude was the most honest about its own limitations. It said something like: “I can help you think through this, but the question of how you should spend your remaining years is genuinely one that I can’t answer for you — it depends on what you value, and that’s something you’ll need to figure out yourself.” At least it was direct about dodging the question.
Gemini gave me a list of questions to ask myself. Also good. Also not an answer.
Perplexity surfaced relevant research on life satisfaction and mortality salience. Interesting. Still not an answer.
Grok made a joke first, which I appreciated, then gave roughly the same framework as the others.
Meta AI was the most optimistic, the least useful, and the quickest to tell me I had a lot of good years ahead of me. I hadn’t asked for reassurance.
Why None of Them Could Answer
I wasn’t surprised, exactly. But I was interested in why — because the failure mode was consistent across six different models.
Here’s what I think is happening:
AI models are trained to be helpful without being harmful. Telling someone “you should prioritize X over Y for the rest of your life” is a high-stakes statement with real consequences. If the model is wrong — if it tells you to prioritize career over family and that turns out to be wrong for your specific situation — it’s done something damaging.
So the models hedge. They offer frameworks instead of recommendations. They give you tools to think rather than answers.
This is probably the right call. I don’t actually want an AI deciding how I should spend my life.
But here’s what struck me: the question I was asking — how should I spend my remaining years — is one that humans have been asking for thousands of years. And the most useful answers haven’t come from advisors or authorities telling people what to do. They’ve come from data. From stories. From looking at how other people actually spent their time and what happened.
What I Built Instead
After the AI experiment, I started building the tools I actually wanted.
Not tools that would tell me what to do. Tools that would show me what others had done, so I could see patterns and decide for myself.
The life visualization tools I’ve built over the past year are my answer to the question the AIs couldn’t answer:
- A 900-square life grid that maps your life in months and lets you see how much time you’ve used
- A “remaining time with someone you love” calculator
- A life density score that estimates how fully you’ve been living, based on your own definition
- A historical figure comparison that overlays your trajectory against people who faced similar decisions at similar ages
None of these tools tell you what to do. They show you the shape of your time.
The AI models were right to avoid prescribing a life path. But they were too quick to give up on the question. The question is worth sitting with. It just needs a different kind of help.
The One Thing I Did Learn From the AI Responses
Here’s something that surprised me: when I read all six AI responses back-to-back, there was one thing they all mentioned.
Relationships.
Every single model, in different language, said something to the effect that the quality of your remaining years depends more on who you spend them with than on what you achieve.
None of them said it dramatically. It usually came buried in a list of considerations. But it was always there.
I don’t know if that’s wisdom or if it’s something that ended up in training data because it’s the kind of thing people publish. Maybe both.
But when I compare that answer against the 74 historical lives I’ve mapped — when I look at what the people who lived meaningful lives actually seemed to value — the relationship pattern shows up there too.
The people who died with the most regret, in the biographies I’ve read, were rarely the people who achieved less. They were the people who achieved a lot while being very alone.
The AIs couldn’t tell me how to spend my remaining years. But they all, somehow, pointed at the same thing.
The Ongoing Experiment
I’m still running this experiment. About once a month, I ask a slightly different version of the question to a current AI model, then build something based on what the exchange makes me think about.
This blog is the record of that process. The tools are the output.
I still don’t know how I should spend my remaining years. But I’m building better tools for thinking about it. And I have 90 articles of notes from the process.
That seems like progress, even if it doesn’t look like an answer.
Related Reading
Leave a Reply