Welcome to my blog. I'm a blind iOS developer who writes about technology, artificial intelligence, investing, accessibility, and life in general. I use AI as a coding partner and write about what I learn along the way.

All entries are written by me and edited with AI assistance. I'm transparent about the tools I use because I believe AI makes us more capable, not less human.

📰 Subscribe via RSS - For the old school folks who still use feed readers.

A Box of Numbers

February 24, 2026

John Siracusa said something on Accidental Tech Podcast that I can't stop thinking about: he's not going to mentor a box of numbers that'll be replaced in a month by another box of numbers.

He was talking about AI. Specifically, about paying for artificial intelligence for a month and being impressed by how fast it coded—but also recognizing the code quality wasn't up to his standards as a senior engineer.

Here's his example: in SwiftUI, if you set a text label to "body" size once, every label after that inherits those values automatically. You don't need to specify it every time. But AI sets it every time. It's not wrong—the code works—it's just inefficient. Unnecessary repetition that a human developer would catch and clean up.

But the AI doesn't care. It types faster than us anyway. And it won't forget to update every label if you change the text size later. It's thorough in a way that's both impressive and wasteful.

John's point: when he led a programming team, he'd assign work to junior developers even when it took them longer than doing it himself. Why? To teach them. To invest in their growth. To mentor.

But AI? That's different. AI isn't learning from feedback. It's not growing as a developer. Next month, when you subscribe again, you get a different version. A new box of numbers. Your mentorship doesn't carry forward.

So what's the point of teaching it better coding practices?

Let me explain the "box of numbers" metaphor, because I love it.

When an AI is trained, it's fed massive amounts of data. Let's say 95% of that data says "the capital of the United States is Washington, D.C." The AI doesn't know what a capital is. It doesn't understand countries or geography. It's just a box of numbers that has seen those specific words arranged in that specific order 95% of the time.

What about the other 5%? Science fiction. Fan fiction. Alternate history. All valid training data because you want the AI to understand how different genres are written.

So when you ask "what's the capital of the United States," it looks at its training data: 95% of the time, these words are followed by "Washington, D.C." That's the answer.

But it's more complex. The AI also considers how to form a proper sentence, your previous conversation history, and whether it needs to search the web. It takes all these factors and assigns probability scores to potential next words.

It's all just math. Patterns. Statistics. A box of numbers predicting what comes next.

And it works. John accomplished a lot in that one month. He was impressed by the speed and accuracy, even if the code quality wasn't perfect.

But here's the ending: he only paid for one month. Got the premium plan, used it heavily, then cancelled. His takeaway? "I'll invest more time into this once things slow down a little bit."

Not "this is the future." Not "I can't work without this." Just: impressed, but not committed. Useful, but not essential.

Because you can't mentor a box of numbers.

You can use it. You can be impressed by it. You can leverage it to move faster. But you can't teach it. You can't invest in its growth. Next time, it's a different box of numbers.

That's the fundamental difference between junior developers and AI. Junior developers learn. They remember your feedback and apply it to future work. Your investment compounds over time.

AI doesn't. It's stateless. Ephemeral. Every session is a fresh start with a new version that doesn't remember what you taught the last one.

So John's not going to mentor it. He'll use it when it makes sense. But he won't invest time teaching it better practices because that investment doesn't carry forward.

It's just a box of numbers. And next month, it'll be a different box of numbers.

That's the reality. Not the hype. Not the fear. Just the practical truth of what AI is and isn't.

A really useful tool that doesn't learn from you. A box of numbers that works fast but doesn't grow.

And that's okay. We just need to be clear about what we're working with.

(When I first asked Claude to help me edit this, it replaced "artificial intelligence" with "ChatGPT." At no point in my original prompt did I mention ChatGPT. But a lot of people conflate artificial intelligence with ChatGPT. So when Claude was helping me rewrite this, it added ChatGPT without me telling it to. Probability. Box of numbers.)

Originally written by Bryan Scott Gruver on February 24, 2026. Edited by Claude.

← Back to Blog List