The father of Node.js officially announced: The era of human handwritten code is truly over.

Ryan Dahl, the father of Node.js, tweeted today and announced that the era of humans writing code by hand has come to an end! Meanwhile, Emad Mostaque, the founder of Stability AI, predicted that the top - tier AI programming experience, which currently costs $200 per month, will drop to less than $1 in two years.

Ryan Dahl, the father of Node.js, posted a tweet today that really stirred up the programmer community:

The era of humans manually writing code has ended!

This post received nearly 1 million views. The big shots left comments in the comment section, expressing their agreement.

And AI - written code will become incredibly cheap, cheaper than you can imagine!

In two years, the top - tier AI programming experience that currently costs $200 per month will drop to less than $1!

This isn't a wild guess. It's the judgment just given by Emad Mostaque, the founder of Stability AI.

His logic is: “The crash of Thinking Tokens is imminent.”

A major upheaval in AI programming is approaching.

1000:1, the astonishing cost of AI programming “thinking”

How much does it cost to get AI to write a line of production - level code for you?

The answer might exceed your imagination.

Let's first clarify the definitions:

Thinking Tokens: Refers to the reasoning process that AI goes through before outputting the final answer.

Code Tokens: Refers to the actual code that is finally written.

Emad Mostaque gave a figure: 100 to 1000 to 1.

That means for AI to write 1 high - quality “Code Token”, it needs to consume 100 to 1000 “Thinking Tokens” in the background.

What does this figure mean?

When AI is writing code for you, more than 99% of the computing power is spent on “thinking”.

You may think AI is a genius that can write code in a second, but in fact, it might have thought thousands of times in the background before giving you an answer.

This is why, even though AI programming tools like Claude Code and Cursor are so powerful, they still require a huge amount of computing power and expensive subscription costs in the background —

Because this “expensive thinking” is the real major cost.

According to OpenAI's official documentation, although the “Thinking Tokens” of the inference model are an internal processing process and are not shown to users, they are charged according to the output tokens. That is to say, all of these are the user's usage costs.

Research also shows that “bad smells” in code can significantly increase token consumption.

That is to say, if your programming taste is poor or your Promt is not well - written, others will consume far less than you in AI programming under the same requirements.

The cost of AI programming is plummeting at a ten - fold speed

But the good news is that the cost of this thinking and reasoning is dropping at a speed of more than ten times per year.

Emad Mostaque predicted that with the decline of computing power costs and algorithm optimization, this “expensive thinking” is becoming extremely cheap.

Even more explosive is his next judgment: This “1000:1” ratio will soon completely collapse.

Within two years, most code will no longer require AI to go through a long reasoning process, but can be directly and accurately generated through One - shot (a single prompt) or Few - shot (a few prompts).

In other words, AI programming will evolve from “thinking - type” to “reflex - type”, that is, AI doesn't need to think and just gives the answer directly. (It's very similar to the experience from Gemini 3 Pro to Gemini 3 Flash)

This means that the price of AI programming is about to plummet. Emad Mostaque directly predicted:

The top - notch programming AI experience worth $200 per month now (such as Claude Code + Opus 4.5), will cost less than $1 per month in two years.

From $200 to $1, a 200 - fold drop.

AI programming is about to enter the “pennies era”.

Data support: According to the research of Epoch AI, the median decline rate of AI inference cost is 50 times per year, and after January 2024, the fastest decline rate even reached 900 times per year!

Taking the performance of the GPT - 3.5 level as an example, from November 2022 to October 2024, the cost has dropped by more than 280 times.

The analysis of a16z also points out that this decline rate far exceeds the historical growth rates of Moore's Law and network bandwidth.

“The crash of Thinking Tokens is imminent”

As soon as the news came out, netizens were in an uproar.

Some people joked: “Thinking Tokens are about to rug (crash)!”

This meme from the crypto circle implies a cruel fact:

The model that relies on a large amount of reasoning to obtain results will soon be replaced by more efficient technologies.

Some programmers also showed off their achievements using AI and sighed at the terrifying speed of efficiency improvement.

Some people even began to worry: If the cost of AI programming drops to a penny level, what's the value of programmers?

When everyone can write code, where is the advantage of developers?

Emad's answer is: Create things that people really need.

Data support: According to the experimental research of GitHub Copilot, developers using AI programming tools can improve their task - completion speed by 55%; users of Amazon Code Whisperer reported that the average task - completion speed increased by 57%.

The Stack Overflow 2025 Developer Survey shows that 84% of developers are using or planning to use AI tools, and 51% of professional developers use them every day.

The Epoch AI 2025 Impact Report: AI capabilities are accelerating

It's not just Emad Mostaque who is talking about the accelerated evolution of AI capabilities.

Epoch AI, one of the world's most authoritative AI research institutions, provided more hardcore data in its just - released 2025 Annual Impact Report.

Report link: https://epoch.ai/blog/epoch-impact-report-2025

First, the cost of AI inference is dropping rapidly.

The analysis of inference price efficiency in Epoch AI's Data Insights shows that the cost of AI inference is being optimized at an astonishing speed.

This is completely consistent with Emad Mostaque's judgment.

Second, the inflection point of the acceleration of AI capabilities may have appeared.

The “Epoch Capability Index” (ECI) launched by Epoch AI is a composite indicator that combines dozens of benchmark tests.

Through ECI, they identified that there might have been an acceleration of AI capabilities around April 2024.

The next acceleration point is about to appear, such as the upcoming DeepSeek - V4 and the latest models of OpenAI and Google in 2026.

Third, long - range software development capabilities are being tested.

The long - range software development benchmark test that Epoch AI is developing in cooperation with METR aims to evaluate the ability of AI systems to complete complete software projects, rather than just fixing bugs or optimizing code.

All these signals point in the same direction: The explosion period of AI capabilities may come faster than we think.

Data support: Due to the authority of Epoch AI, its website has 987,000 active users (a 4.3 - fold increase year - on - year) and is cited by 107 well - known media and reports (a two - fold increase year - on - year). It also provides data and consulting services for top - tier institutions including OpenAI, Google DeepMind, and the US Congress.

AI inference is becoming “dirt - cheap”

Currently, the inference cost of top - tier AI models ranges from a few dollars to dozens of dollars per million Tokens.

The API cost of the GPT - 5.2 series:

The API cost of the Claude series: