SuperMuse Blog

AI Did the Work. What Did You Learn?

Yiming — Tue, 15 Sep 2026 04:00:00 GMT

AI can write code, explain a paper, and quickly put together a working tool. As results get easier to obtain, it’s natural to wonder whether we still need to spend so much time learning. I’m interested in what happens afterward: when a similar problem comes up again, will we understand it any better?

In a 2023 interview, CNBC’s David Faber asked Elon Musk what career advice he could offer children entering the workforce as AI improved. Musk paused, acknowledged the difficulty, and suggested pursuing interesting, fulfilling work that was useful to society. The uncertainty extends to learning: will the time we invest today still be worth something tomorrow?

I recently read a short post by Simon Willison in SuperMuse about three blog posts that had influenced him. One was Joel Spolsky’s 2002 essay, The Law of Leaky Abstractions. I’ve long enjoyed Joel’s writing and had read this one years ago. Simon’s post sent me back to it.

Reading Joel Again

Simon said Joel’s blog post encouraged him to keep learning about the layers beneath his work, in case an abstraction leaked.

An abstraction gives us a simpler way to work with something complicated.

Objects bundle state and behavior behind methods, hiding their internal implementation. Relational databases let us work with tables and queries without managing every detail of file storage and disk access. Object-relational mappers let us work with those databases through objects in our code, handling the mapping to tables and generating common SQL queries. But the hidden details can still affect the result. When they do, the abstraction “leaks.”

When the Details Come Back

Why isn’t one emoji always one character long?

A username field seems easy to limit: check the string’s length. Yet in JavaScript, "😄".length is 2. The property counts UTF-16 code units, not the characters a person sees. Compound emoji can involve even more code units. MDN explains the distinction.

If a counter is wrong or truncation breaks a character, we need to understand encoding and character boundaries. A property named “length” hasn’t settled what a user means by “one character.”

Why can a few lines of database code make a page slow?

An object-relational mapper lets us access records much like ordinary objects. Fetch ten books, then read each book’s author: straightforward enough. Without eager loading, however, that may mean one query for the books and ten more for their authors. This is the N+1 problem, illustrated in the Rails guide.

The framework saves us from writing SQL by hand. It doesn’t remove the cost of database round trips. To investigate the slowdown, we need to see what queries actually run. The few lines on the screen may hide most of the work.

Why isn’t retrying always the answer to a timeout?

Suppose an application sends a request to create an order but never receives a response. The server might not have received the request. Or it might have created the order and lost the response on the way back. Without protection against duplicates, trying again could create a second order.

We need to understand what each side knows and how to retry one operation without repeating its effect. Stripe’s documentation on idempotent requests describes one approach. A convenient function call still depends on uncertain communication.

Clicking Retry is easy. Understanding why it created another order means looking beneath the interface.

These situations share a pattern: to understand what’s happening, we often have to look one layer deeper.

Joel puts it this way:

“So the abstractions save us time working, but they don’t save us time learning.”

The tool removes repetitive work. Knowing how to call its interface doesn’t mean we understand what happens underneath.

A Higher Abstraction Still Leaks

From the user’s perspective, AI-assisted programming adds another layer of abstraction. We describe what we want in natural language; AI translates that into a programming language, and frameworks and systems execute it. A request as simple as “build an order page” can leave AI making decisions about data structures, queries, and request handling.

A higher abstraction can still leak. A page that loads successfully may slow down as its data grows or create duplicate orders when requests arrive together. Once those details affect the result, we have to look at the code, database, and communication between systems. Without some understanding of the underlying mechanisms and experience in software engineering, it can be hard to locate the problem or judge whether a fix is sound.

In this sense, Joel’s law of leaky abstractions still applies to AI-assisted programming. Natural language makes it easier to begin; the rules governing the system remain. Understanding the underlying system helps us know what to question—and how to check it.

This new layer can make us more productive without giving us a corresponding understanding of the work. Getting things done faster doesn’t mean learning happens automatically, or becomes easy. AI can explain mechanisms and help devise experiments, but learning them thoroughly still takes reasoning, practice, and revising our judgments in response to feedback.

A Correct Answer Can Leave Us No Wiser

There’s a subtler possibility: the AI is correct, the task succeeds, and we learn very little.

In an interview this August, Andrew Ng discussed cognitive offloading: delegating some thinking can also remove opportunities to practise. He described asking AI how frontend or backend components worked, finishing the project, then facing a similar task six months later without remembering the answer. He had to ask again.

That isn’t evidence that AI damaged his memory. The more immediate concern is that the knowledge may never have been thoroughly understood or retained.

A study in PNAS tested AI-assisted maths learning with nearly a thousand students at a Turkish high school. Students using a standard GPT-4 chat interface did better during assisted practice, but scored about 17% below the no-AI control group on subsequent independent exams. A tutor designed to offer teacher-informed hints and withhold direct answers largely avoided that harm, though it did not significantly outperform the control group on the independent exams.

One study doesn’t cover every learning situation. But it makes an important distinction concrete: performance on a task and learning from it are different outcomes.

A clear explanation and a successful attempt can feel like understanding. A changed condition—or the need to explain the idea from scratch—can reveal what we’re still missing.

Our End of the Lever

These days, AI is often described as a lever. It lets one person do work that once required more time and more people. At our end of that lever is what we bring to it.

As the lever grows longer, the weight at our end still matters. The more understanding we bring, the heavier the load we can lift.

That weight includes knowledge, experience, and familiarity with the problem. We might question whether an article’s evidence supports its conclusion, or distinguish a user’s actual difficulty from a need we’ve imagined. Those insights help us ask AI more useful questions and turn a first answer into something worth using.

Experience doesn’t guarantee good judgment, and beginners can use AI to accomplish things they couldn’t before. We can add to the weight at our end as we go. Each thing we understand a little better gives us more to try next time.

Two Extremes to Avoid

Taking learning seriously can turn into an insistence on doing everything the hard way: every line of code must be handwritten for the work to count. But understanding a network request doesn’t require implementing the entire protocol stack from scratch. Tools can spare us repetitive work and leave more time for the questions worth studying.

The other extreme is outsourcing everything, including judgment and verification. If the result runs, we stop asking why; if it fails, we keep asking AI to try again. We may finish plenty of tasks without becoming better prepared to investigate when an abstraction leaks.

I’d rather use the tools and learn as I go. Let AI help with the work while keeping opportunities to question, verify, and try things independently. The problems in front of us can guide what we need to understand next.

See the Whole, Then Explore the Details

Leaky abstractions give us a reason to understand what lies underneath. But we can also lose the forest for the trees. Start with the problem we’re trying to solve and how its concepts fit together; that helps us decide which details deserve attention. Once we understand a detail, step back and ask where it belongs in the larger picture.

AI can help us build that picture: outlining a subject’s central ideas, comparing similar problems across fields, and connecting new knowledge to what we already know. Treat this as an initial map, then check and revise its connections through original sources and experiments. Our notes can record how our understanding changes along the way. Learning should leave us something to think with next time.

Reading with Muse

In SuperMuse, AI Muse can accompany that process. Discuss a passage together, then let it test your reasoning with a question, a counterexample, or a scenario with a tempting wrong answer. Having to explain a choice can reveal what we haven’t quite worked out.

After reading, bring new highlights and old notes together with Muse, and decide which connections hold up. There’s a Chinese phrase I like: 温故而知新—finding new understanding by revisiting what we’ve learned. A new question can give an old note fresh meaning. Keeping reading, notes, and discussion together gives us a chance to make that knowledge our own.

AI can help us get the work done. I’d like us to come away understanding a little more, too.

References and Further Reading

Simon Willison: What blog posts influenced your thinking the most? (September 14, 2026)
Joel Spolsky: The Law of Leaky Abstractions (November 11, 2002)
Will Larson: Migrations: the sole scalable fix to tech debt
Charity Majors: The Engineer/Manager Pendulum
CNBC: Elon Musk on his children’s future in the workforce (May 16, 2023); contemporary transcript
Silicon Valley Girl: Andrew Ng: The Biggest Opportunities in AI Aren’t Where You Think (August 28, 2026; learning discussion begins at 11:59); transcript
Bastani et al.: Generative AI without guardrails can harm learning: Evidence from high school mathematics (PNAS, 2025)
Oracle: Objects, encapsulation, and implementation details
PostgreSQL: Relational database concepts
Rails: Object-relational mapping
Rails: N+1 queries
Stripe: Idempotent requests
MDN: String length and UTF-16 code units

Why SuperMuse?

Yiming — Sun, 26 Jul 2026 01:30:00 GMT

Before it was SuperMuse, it was VoxMind. I had a fairly simple idea: an AI-native tool for taking notes and developing ideas, built around voice. A thought rarely arrives fully formed. Sometimes it’s just a sentence that comes to you on a walk, in the car, or over coffee. I wanted to be able to say it out loud, without waiting to get back to my computer, and have AI help me capture it and think it through. Those fragments could eventually become an article or something else.

But the longer I worked on it, the less convinced I was that helping people write faster with AI was what I really wanted to do.

AI is already good at generating content. Give it a prompt and it can write an article or generate an image, a piece of music, even a video. It’s getting faster, and the results are improving. But as producing something becomes easier, I find myself wondering whether the hardest part comes before we start: Why do we want to write about this in the first place? Where does an idea of our own come from?

Muse

A writer might call someone “my muse”; a photographer might say the same about a city. We often use the word to mean a source of inspiration.

In the best-known Greek tradition, the Muses are nine goddesses of art and knowledge. Poets would often open by asking a Muse to help them tell their story, as Homer does in the Odyssey.

A poet listens to the nine Muses.

I like this view of creative work: the poet draws on what they’ve heard and remembered, then tells the story in their own way.

Memory Gave Birth to the Muses

To me, the word Muse suggests memory, discovery, and connection, as well as creativity.

But what really drew me to the name was learning that the nine Muses were daughters of Mnemosyne, the Greek goddess of memory.

Mnemosyne: fragments of memory becoming inspiration.

In Hesiod’s Theogony, Zeus and Mnemosyne spend nine nights together, and she later gives birth to the nine Muses. I love the metaphor in that story:

Memory gave birth to the Muses.

Inspiration, born of memory.

A book we read years ago can suddenly help us understand a problem we’re facing today. An idea we once shelved might become possible thanks to new technology. Sometimes that flash of insight comes when something from our past connects with what we’re thinking about now.

Memory → Connection → Inspiration

A new question brings separate experiences together.

Inspiration Is Only the Beginning

Most of my ideas start out vague. A product might begin as a single sentence; an article, as a feeling that there’s something worth thinking through. I read, talk it over with AI, and make a first version. That usually reveals something that doesn’t quite work, and I go back and rethink it. Sometimes I have to start making something before I understand what I want to make.

We don’t simply put ideas into practice. We shape and refine them through the work itself.

Context Is More Than Memory

Our context includes the article we’ve just read, the draft we’re working on, and what we’ve learned from the last attempt, alongside our memories. It changes as the work develops.

That’s what interests me about context engineering: What does this idea need right now? When I’m stuck, an example might help. When I’m too sure of myself, I might need a counterexample. What we bring into the discussion can change where our thinking goes next.

As we give AI the context it needs, AI can help us put our own thinking in a new context.

Each attempt teaches us something to bring back to the work.

AI Can Be a Muse

I’d like AI to be a Muse: to spark ideas and question my assumptions. I still decide what to keep, what I agree with, and how to say it. After a good discussion, I should have a clearer sense of what I think.

The creator is still human.

The Muse is there to help. The work is still ours to make.

SuperMuse

VoxMind made sense when the idea was to capture thoughts through voice. As I kept building, I wanted the product to help those thoughts take shape, too. SuperMuse felt closer to what I had in mind.

Adding Super was mostly a practical choice. SuperMuse was easy to remember, and Muse on its own was hard to register.

If there’s a hope behind the name now, it’s that this Muse can help us stay with an idea a little longer, while we’re still figuring out what it could become.

Why I Built SuperMuse

Yiming — Sat, 27 Jun 2026 01:30:00 GMT

A day by the salt ponds

By the pool

Among the redwoods

Facing the ocean

An August visit to Google

Out for a ride

A cable car in the city

We are ebay

Time for a rest

These nine cards were made with SuperMuse, pairing photos from travel and everyday life with dates, places, or a few words to remember them by. Swipe through the cards, or select one for a closer look.

I’ve got a few hundred half-finished ideas in my notes app.

I save passages from books and jot down product ideas when I’m out for a walk. There are story openings, coding problems I haven’t figured out yet, and photos from trips with a few words about the moment. When I save them, I usually think they might come in handy later.

But when I need something for a piece I’m writing or a question I’m trying to work through, I often can’t find it.

Sometimes I remember an argument from a book but can’t remember which book. I try a few searches and get nowhere. Other times I come across an old note and realize it’s relevant to something I’m thinking about now. I hadn’t seen the connection when I wrote it.

Connecting the dots

In his 2005 Stanford commencement address, Steve Jobs said:

“You can only connect them looking backwards.”

In that story about connecting the dots, he recalled how a calligraphy class he attended in college later influenced the typography of the Macintosh. He was talking about his life, but it makes me think about the things we jot down. Something we save without knowing what it’s for can turn out to be useful later.

Folders and tags help me organize things. They can’t tell me what a note might connect with in the future. I wanted to build SuperMuse to keep my reading notes, photos, and unfinished ideas in one place, so I could find what I needed when a question came up and keep thinking it through.

Some conversations might lead to an article. Others might add just a few sentences to a note. Being able to pick up an unfinished thought is useful in itself.

Capture the thought first

When an idea comes to me, I don’t usually have a title for it, let alone a folder in mind. If I have to organize it before I can save it, I might not bother. I should be able to say a few words, jot something down, or save a photo or link and leave it at that.

I also want to keep a little of what I was thinking at the time. Beside a quotation, I might write what I don’t understand. With a photo, I might add what it reminded me of. That gives me something to come back to, beyond the passage or image itself.

I can work out what it’s for later.

Bring your notes into the conversation

Sometimes I agree with an author; sometimes I think of a counterexample. I want a reader that lets me keep the passage and my questions together. If a concept isn’t clear, I can ask AI Muse for an example. Or I can explain my understanding and ask what I’m missing, then go back to the book with new questions.

Say I’m reading about negative space in design. It might remind me of a poster I photographed on a trip and a product decision I’m wrestling with: would adding more features actually help? I could bring that passage, the photo, and my own notes into a conversation with the Muse. Is leaving space on a poster really the same as leaving features out of a product?

Maybe I’ve been treating two different things as the same. If an explanation doesn’t hold up, I can go back to the passage, try another example, and revise my note.

I can do something similar when I’m stuck on a coding problem: bring in a technical article and talk through the tradeoffs between a few approaches. I still need to check facts and quotations, and see how the ideas hold up in practice.

That’s what AI-native means to me. AI is there as I read, take notes, and create, helping me ask questions and revise my thinking as I go. I don’t want to switch windows and explain everything from scratch whenever I get stuck. I want to keep going with the material in front of me and choose which of it to bring into the conversation.

Write down what you’ve worked out

Once I’ve worked out what I think, the Muse can help me turn my notes and examples into a draft. I can explain what first puzzled me, where I’ve landed, and what I’m still unsure about. I can ask it to point out what’s hard to follow. Then I can fill in the gaps and revise until the piece says what I mean.

For a product idea, that might mean talking through something I can try, making a first attempt, and bringing what I learn back into the next discussion.

Other things only need a photo and a few words. The nine cards above pair photos from travel and everyday life with a date, a place, or something I wanted to remember. Like my reading notes, they’re things I’d like to come back to.

Not every note needs to become a finished piece. Some are just for me. Others might find their way into an article much later.

I’m building SuperMuse because I want to be able to use what I’ve already saved. The next time a question comes up, I’d like to find that passage, revisit an old photo, or pick up a note I barely started.

Explore SuperMuse →