What Still Matters When AI Gets Better

I've spent years working on systems that listen to conversations, form memories and turn human interaction into structured information. I've shipped hardware that does this. I am interested in the technology. I am also tired of the feeling that missing a week of news means falling out of the world.

Underneath that feeling is a question I have been circling for months: if more of what I can do becomes cheap to reproduce, what should I spend my life getting better at?

I do not have a list of skills that machines will never acquire. I am suspicious of that kind of insurance. A claim about what today's system cannot do has an unfortunate habit of becoming next year's demo.

Knowing enough to disagree

Access to an answer and understanding the answer are different achievements. AI makes the first easier. It can help with the second, too, but only if I stay involved enough to notice when an explanation does not add up.

Knowing things still matters. It lets me ask a less obvious question, catch a missing assumption, or recognise that a confident answer has used the wrong unit. Without that grounding, I am dependent on the same system for the answer and for the judgment that the answer is good.

So I want to use these tools to learn, including the slow parts: working through a derivation, predicting what a program will do before running it, explaining an idea without the generated summary open beside me. The convenience is useful. I do not want it to cost me the ability to check.

The parts with consequences

A difficult conversation teaches something that a description of difficult conversations does not. You have to choose the words, watch how they land and live with what follows. Building something exposes your assumptions to people who have no obligation to agree with them.

These experiences matter to me without requiring a proof that machines can never learn from experience. I care about becoming a person who can exercise judgment when I cannot outsource the consequences. That includes knowing when to rely on a tool and when to slow down.

The same applies to relationships. A well-written message can help me say something. It cannot stand in for the years in which I did or did not show up. I want the people close to me to have evidence of my attention in their own lives, beyond what I am capable of writing about it.

Work and identity

Advaita offers a distinction I return to: the activity of thought and the awareness in which it is experienced. I find that useful when a change in my work starts to feel like a threat to who I am.

I do not need to turn that into a scientific claim that AI cannot have experience. We do not have an agreed test that settles machine consciousness. What I can examine directly is my own tendency to equate my worth with being useful, fast or technically ahead.

That tendency was a problem before AI. Better models make it harder to ignore. If I build my whole identity around outperforming a tool, I will be checking a benchmark to find out how to feel about myself.

A smaller set of commitments

For now, I want to keep a few things in view:

• Learn enough about the work to question an answer, even when producing the answer becomes easy.
• Make time for people without turning that time into another measure of productivity.
• Use silence, reading and physical activity to notice what constant input makes difficult to hear.
• Build things whose consequences I am willing to take responsibility for.

None of this guarantees a safe career. There are real changes in employment and bargaining power that a personal philosophy cannot wish away. But it gives me a way to choose what to practise while the technology changes.

I will keep using AI, and I will keep missing some announcements. I would rather miss a model release than spend another year unable to say what I was paying attention for.