Learning has been my way of trusting the future

Recently, I suggested that my fiancée use an AI agent for some manual research work she had been doing for weeks (something like filling forms in a manual, mundane way). She is the smarter of the two (meaning me and her). She is a wonderful researcher, currently doing her residency in ophthalmology, and someone who has studied and used AI herself. Still, the suggestion that an agent could literally operate her computer and handle this work did not initially sound serious to her.

I understood her reaction. Using these tools regularly, I still find myself revising what I think they can do. With the latest GPT-6 Astra, seeing how much more these tools can do leaves me feeling excitement and discomfort at the same time.

Generally speaking, I see that AI agents (meaning Claude Code, Codex, and such) make me more willing to attempt things. I can explore an unfamiliar area, build something, or fix a computer problem with less hesitation, at least benchmarking against my past self. I already consider myself open to exploration, especially anything regarding tech and computers (yes, I'm a physician, thank you for reminding me, but I know my way around tech a bit. I started following MKBHD, my man, back in the days when he had only a couple of thousand subscribers, in his college apartment). Anyway, my past self, although explorative, would never even have attempted to analyze data about the linguistic abilities of LLMs vs humans, for example. AI agents let me explore more, think more broadly about my capabilities, and actually do more. Saying they save me time is a stretch, though, and I discussed that in my earlier essay about saved time.

Yet, I believe that part of who I am is built around capabilities that took years to develop, and the process of learning, trying, failing, and trying again to achieve them. Personally speaking now, if I had to name the trait that helped me advance most, I would probably choose my ability to learn. And I think the ability to learn and change is one of the innate most humane capiabilities, that make us continue to exist, with Allah’s willing, as a specices. Whatever changed, I trusted that I could learn enough to find a way forward. In essence:

Learning has been my way of trusting the future.

So seeing AI become more capable raises a rather personal question (and I think many people feel and ask the same, in one way or another): what happens to that trust if much of what I could learn becomes available on demand? What happens if we get to the point where there is even no practical benefit to learning a new skill or ability, because an AI agent can do it in an instant and walk you through it? Even if my professional position were secure (money included), I think some discomfort would remain.

Lets consider this, for the matter of making a point: we often tend to judge capabilities through visible outputs. Take this example, rather relevant at the time of writing: a beautiful, accurate 3D model is easy to appreciate. Dependable help across an ordinary working week is harder to demonstrate. So if one AI agent can produce the first better, while another can do the second better, the first agent will probably be the “social media cahmp!.” I wonder whether I judge myself similarly, giving more weight to what I can produce than to everything that happens while I become able to produce it (and I wonder about us collectively, too).

Going back a bit, actually, this idea of expansion in capability has some empirical support. In an experiment involving 791 P&G professionals, individuals using AI matched teams without AI on the quality of product-innovation proposals. AI also helped participants cross their usual technical or commercial specializations. This was a particular task, but it illustrates why the excitement feels justified: work that once required a broader team can become accessible to one person. Organization Science, 2026

Our relationship with the resulting work deserves attention too. A writing-task experiment, with 269 participants in its main analysis, found that people instructed to copy AI output reported less ownership and meaning than those who wrote independently or drafted before using AI to refine. Although these were short-term, self-reported experiences, I really do see them as suggesting that how we participate can matter alongside what gets completed. Scientific Reports, 2026

So, in a way, learning often gives me things that a finished output cannot show. Experimenting and getting feedback can help me judge a result, recognize a weak assumption, or notice that I have been pursuing the wrong question. Those abilities feel particularly valuable amid impressive AI demonstrations and uncertain claims.

But even that defense makes learning justify itself through better future performance (right?). For me, some of its value is already there while it happens. I enjoy understanding something I did not understand yesterday. I enjoy experimenting without knowing exactly where it will lead. If I wanted only to know how a Sanderson book ends, I could ask for a summary and save myself quite a few pages. I suspect I would have missed a fair bit of why I picked it up.

“Journey before destination,” from The Stormlight Archive (shoutout to my man Brandon Sanderson! Though he does not know I exist), captures something I find easy to agree with and easy to overlook when evaluating my own work. I suspect many others do, too.

If a project is enjoyable, I want to spend time on it. Yes, don’t get me wrong, I would happily ask AI to handle the tedious parts. Removing those obstacles can give me more room for the experience I wanted in the first place. Take this essay: I am happy to let AI handle the Word formatting, grammar and general flow. I would still like to be involved in figuring out what I am trying to say (which is taking a little longer).

To be honest with myself and you, future systems might complicate this choice (just building on trends, I think the pace at which AI has developed and advanced makes even some currently unimaginable future capabilities a fair bet). Perhaps an agent will eventually even anticipate what I want and produce it before I ask. That could be welcome in many parts of life. But when I begin a personal project, activity, experiment, or whatever, my initial intention is often incomplete. Working on it can change what interests me and what I want to make (by the way, I see this essay, and others, as an example of this. GPT-6 with my agent skills can write in a much more organized way and get straight to the point compared with my own writing, at least with some trial and error. But still, I write it myself, because the process changes as it happens. I get new ideas, throw them in, write, and try to convey them. I think these essays are more an expansion of my thoughts, unresolved and resolved, than just a transfer of knowledge. And I find this way of writing, for example, really hard for an AI agent to grasp).

So, simply (or not), a system that perfectly fulfills my starting intention might miss the intention I would have developed along the way.

I hope increasingly capable AI will help with that discovery too, introducing possibilities I would not have encountered alone and giving me room to choose among them. We could evaluate these tools partly by whether using them helps us explore and develop our interests. Faster completion would still matter, alongside what the experience makes possible for us.

I remain excited about becoming more capable with AI. And at the same exact time, I also remain uneasy and discomforted, because I KNOW that I have yet to perceive the future with AI as it will eventually be, and my role (and ours) in it. But, one way or another, I also suspect that learning will remain part of how I understand myself, even as its practical role changes.

An agent may eventually know what I want before I do. I would still like some room to discover that I want something different.