Working with AI in 2026

We're at a tipping point when it comes to AI. I've been using these tools for a long while now, but recently, with the release of Claude Opus 4.6, I think it's really over. Before, Claude was just a helper. Now, it's everything. Before, it would make mistakes or leave something unfinished. Now, it always gets the job done and even cleans up after itself. It's really got me thinking about the job market and the future of labour.

I had a conversation with a friend about how most of the Quran is just 500 words. That is to say, if you understood the meanings of 500 words, you would understand about 75% of the Quran.

I went home and thought that an app for this would be a great solution.

Step 1: Getting the Words

The first thing that I needed to do was to:

get all of the words in the Quran
get each of their frequencies
their transliteration
their translation
a voice speaking these words in Arabic

Unfortunately, there exists no place where that database exists. I found a website which has the words and their frequencies, but only transliterations for some of them. Still, I thought this was better than nothing.

So I got AI to write me a scraper to get all these words and store them in a JSON file.

Okay, but I still need the translations for each word.

So I got AI to write a script which scraped that same website to get the Quranic translation for each word and save it in my JSON file.

Now I need to get the transliteration of each word.

So I got AI to write a script that transliterated an Arabic word to English.

But it didn't take into account some Arabic rules.

So I told AI it wasn't doing that, and it fixed it.

I'm going to do this for each one. Now I need a person saying each word in Arabic.

So I got AI to generate 5,000 voice files, one for each word.

These voice files have a pause before speaking and after.

So I got AI to write a script that would kill the dead air.

Great! Now I have my database. I need to actually create the app, so I write a full spec on exactly what I want for this app. Full of complex features and database management systems. I think Opus 4.6 will be able to do it, but it will have some bugs, or it will not look great, or it will lose something in the context.

But no. Absolutely flawless. Runs on the first try. The home page looks a bit dry.

Claude, put some cool charts and stats on the home page.

Done. Five different charts I didn't even think of. This is a monumental amount of code. This would have taken a skilled dev weeks to finish. I finished it in two hours with Claude. And I looked over the code, and it's actually good—for the most part. Some things I would have done differently, but if somebody told me someone with 10 years of experience wrote it, I would not be shocked.

The scary thing about this isn't that it's so good, but more so how fast the progress seems to be accelerating. 4.5 was a huge leap, but I feel like it just came out, and 4.6 comes along and blows it out of the water. And now they're saying recursive improvement will only accelerate even further. It feels like a whole new paradigm, something fundamentally changed since 4.5 was released, and we can never go back.

A year ago, Claude was just a tool. A year from now, I think it will be the best programmer the world has ever seen. And with the rate of acceleration, I don't think a person can even predict where it will be a decade from now.