# AI Makes Building Easier. I Still Have to Learn.

**Summary:** AI helps me build useful tools and attempt creative work, but I still need to learn how to explain and judge the results. A reflection on mistakes, understanding, and personal usefulness.

- Canonical: https://markhuang.ai/blog/ai-makes-building-easier
- Language: en
- Author: [Mark Huang](https://markhuang.ai/about)
- Published: 2026-11-11
- Section: Behavior & Mindset
- Tags: ai, continuous-learning, generative-ai, self-reflection
- License: https://creativecommons.org/licenses/by-nc/4.0/

---

![A creator studies sketches of the same scene beside a display, comparing the picture with the choices on paper](https://cdn.markhuang.ai/blog/ai-makes-building-easier/hero.webp)

*Illustration: getting an image gives me something to examine. Learning helps me work out what to change.*

In 2026, I have been experimenting with AI-generated videos, images, and other creative work. Seeing people use Claude to make animations makes me want to try it too. I have an idea of something I would like to make, and AI gives me a way to start.

Getting exactly what I want has been much harder. I often do not know which decisions I need to make, or what information I should give the AI. I can see that the result is not quite right without knowing how to explain what would make it better.

That experience keeps bringing me back to learning. AI makes it easier to attempt unfamiliar work. It has not made me automatically understand the work, and I need that understanding if I want more control over the result.

## Something I actually want to use

Part of what excites me about AI is being able to make something I wish existed. Sometimes I want a tool, look around, and cannot find one that fits what I need. Being able to build it for myself is a good reason to use AI.

That is why I built [Dense-Mem](/blog/ai-memory-beyond-rag) and [AI MemMail](/blog/ai-memmail-dense-mem-email-agent). Dense-Mem manages context that AI can remember and recall; AI MemMail brings that context into email handling. I wanted to use these things. I see them helping me understand things better, and I want them to make my life easier.

I am not starting with the question of how to get a popular GitHub repository. I would like other people to find my work useful, of course. Something can be worth building even if I am the only person who finds it useful and it never becomes widely known.

I see a lot of repositories being created and actively promoted. I understand the appeal. People want others to notice what they have made, and a project needs some way to reach the people it might help. With so many tools competing for attention, though, getting noticed and becoming part of somebody's life are different problems.

Stars are also an imperfect signal. [Researchers studying suspected fake-star campaigns on GitHub](https://arxiv.org/abs/2412.13459v2) have documented how that signal can be artificially inflated. That does not tell me which individual project is useful or how many people actually use it. It does make me reluctant to treat the number as a substitute for those questions.

With my own projects, I can start by asking whether I actually want to keep using them. If I do, I already have a reason to work on them, even when nobody else is paying attention.

Creative work starts from a similar wish. I want a particular image, or a video that explains something the way I have it in mind. The difference is that I am much less familiar with the decisions involved.

## I can see the problem before I can explain it

In work I understand, I have some sense of what matters. I can describe a problem, recognize constraints, and decide whether a proposed result is useful. AI can help me move faster because I already have a way to judge what comes back.

With an unfamiliar craft, even the request can hide choices I do not recognize. Asking for a good animation leaves a lot unresolved. Timing and framing, for example, can change how the same action feels. Those are the kinds of things I need to learn enough about to discuss, rather than hoping the word "good" carries the picture in my head.

There is a difference between disliking a result and knowing how to improve it. The first reaction can be immediate. The second asks more of me. If I cannot identify what bothers me, I am asking the AI to guess at something I cannot yet describe.

This difficulty is not unique to me. In a [2023 analysis of more than 100 Midjourney prompts and their resulting images](https://www.nngroup.com/articles/ai-articulation-barrier/), NN/g's Tarun Mugunthan described users struggling to express visual intent, including a lack of vocabulary for the styles they wanted. That was an observation of a particular image-generation tool and its users, but the difficulty is familiar.

Better tools and interfaces can help with that. Clear instructions can also produce disappointing results because the tool has limitations. I am interested in the part I can work on: learning to notice a choice, understand what it changes, and explain it more clearly.

A Chinese saying comes to mind: *人赚不到认知以外的钱*, roughly, "You cannot earn beyond what you understand." I think about it here as an analogy for the limits of my present understanding. Having access to a capable tool does not mean I recognize everything that could be done with it.

That understanding can grow. I can start before I know the whole field, and AI can help me discover concepts I did not know existed. But I still need to spend time with them. Being introduced to a word is the beginning of learning what it means.

## A mistake gives the advice a reason

The amount to learn can feel overwhelming. I see an animation I like, want to try something similar, and then find myself facing a whole set of unfamiliar choices. Before I understand the basics, I cannot reliably separate the details that matter from the ones I can leave alone.

For me, making mistakes and reflecting on them is often the fastest way into that understanding. A failed attempt gives me a specific question to investigate. I have a reason to care about the answer because I have just run into the problem.

Books and tutorials can explain a principle clearly. What takes me longer is understanding the path that led the author there. Which tempting approach did not work? What problem made this piece of advice necessary? I can follow the instruction without yet understanding why someone would have arrived at it.

Trying something gives me a way to connect that explanation to a result. Sometimes the advice becomes much easier to understand after I have made the mistake it addresses. I have a result to compare the explanation with.

Repeating a mistake does not automatically teach me anything, though. If I keep asking AI to try again and accept whichever version looks best, I might finish with a better image while knowing very little about why it improved. The reflection matters: what changed, why might it have helped, and what should I try to understand next?

This is a reason to keep an experiment small. I could compare two versions of a short animation with one deliberate change instead of trying to learn an entire production process at once. That would give me something specific to inspect. It would not require every unfamiliar detail to become today's problem.

AI can be part of that learning. It can suggest alternatives, explain terms, and help me compare examples. A [preprint by Benjamin Lira and colleagues](https://arxiv.org/abs/2502.02880v4) found that people who practiced with an AI writing tool later wrote better cover letters without AI than participants who practiced on their own. The study concerns professional writing, not every skill or every AI tool. It does show why I should not assume that getting AI help prevents learning.

When AI offers an explanation, I want to be able to compare it with the result in front of me. Does it explain the difference between two versions? If it does not, I still have something to investigate.

## What I want to get better at

I have written that I am [aggressively trying to replace myself with AI](/blog/aggressively-trying-to-replace-myself-with-ai). I still want to automate work and make room for things I care about. These creative experiments have made me think more about what I want to do with that room.

In [I Keep Asking Why](/blog/i-keep-asking-why), I wrote about understanding the reasons behind decisions. When I step into an unfamiliar field, there is an earlier difficulty: noticing that there is a decision to make at all. I cannot ask a useful why about a choice I have not recognized.

That gives me a reason to keep learning even as AI becomes more capable. I want useful tools and creative work I am happy with. I also want to understand enough to explain what I like about a result and what I would change.

I expect to keep making things I am not entirely happy with. I want to get better at understanding why, so I have something more useful to tell the AI on the next attempt.
