AGI does not exist

Zeusro Feb 6, 2026

In 2020, when I proposed rebuilding the entire operating system with AI as the gateway (OOOS) in full English, I only treated AI as a "gateway entry" and did not foresee AI programming's dimensional strike against traditional foundational programming.

In 2025, after comparing chatGPT's performance over these five years, I boldly assert that foundational programming roles will largely disappear within five years, and I predicted the birth of Clawdbot.

After using cursor and trae in the second half of the year (trae is slower, only recommended as a backup), I say five years is still too conservative—in two or three years foundational programming roles will be all but gone 🤣.

Preface

2025-04-02 22:33

What various vendors offer today cannot be called an AI agent. The first principle of an AI agent is that core data is stored in an AI model on the user's own device. In the near future, when people realize the importance of their own data (it effectively represents a unique identity on the internet), are unwilling to hand it to traditional internet vendors, and the cost of local AI deployment drops to the level of ordinary digital consumer goods, the AI agent will truly be born. ...... Foundational programming roles will largely disappear within about five years. What are today's programming languages, really? I think they are all a kind of bridge between humans and machines. Machines need to execute this intermediate language, while humans need code to crystallize existing business knowledge. So when today's AI has sufficient access to the current project directory and stronger contextual understanding, in essence AI is replacing this foundational programming work.

For now, however, AI still lacks full file reading and very large context. Current AI collaboration is more like glue—AI completes a small part of the user's needs, and programmers stitch it together. But foundational programming roles are already visibly declining.

Foundational programming work will be largely phased out within five years and enter the next stage of history. By then, a six- or seven-year-old child will only need to drag and drop in something like Swift Playground, enter user requirements, and AI will produce the most efficient implementation. What people will do is use creativity to weave knowledge from various disciplines into program design and meet complex functional needs.

The current version (Version: 2.4.21) of cursor already has full authoring capability at the 100k-line project level. By the 80/20 rule, 80% of projects are 100k-line piles of junk, so 99% of code basically no longer needs to be written by humans.

After finishing my kube-killer with cursor, I fell into Coding ED—I had no desire to write any code at all.

Every idea materialized in an instant, and I didn't even need to spend time learning various business concepts. Over these five years, as the "Crow Seldon" of the Chinese internet, I have already prophesied the decline of several industries. With the rise of Vibe coding, unemployment has finally landed on my doorstep (job offers please send to [email protected]).

Lately I have been thinking about how, in this postmodern society, humans should respond to AI's impact on the current industry landscape. I finally found a way in through formal logic.

AGI Does Not Exist

An omnipotent God cannot create a stone He cannot lift --> AGI does not exist.

Formal logic demands that inference be absolutely correct. But an omnipotent God cannot create a stone He cannot lift. If He could create it, then God would not be omnipotent; if He cannot create it, that only proves God is not omnipotent. By analogy, AGI does not exist. AI has an upper bound on its capabilities.

I believe that bound is formal logic and human emotion. As for the "antinomy" that gave birth to formal logic, AI may never understand it in its lifetime.

Humans need to reclaim their own agency and reject the programming cult of the "cyber sugar daddy" (treating AI as a universal wish-fulfilling machine for every problem).

AI Hybrid Programming Model

Based on the structure of Service As API and Time-Series Metaprogramming, I propose the AI Hybrid Programming Model.

Humans are the stewards of data and the review engineers of AI-generated output. AI must run inside a "sandbox" with restricted execution. AI has read-only access to underlying data; generated artifacts use copy-on-write (copy on write).

Only in small files at small scale can the git protocol be used temporarily. Trade storage redundancy for the safety of underlying data.

The relationship between humans, AI, and data is governed by an Auth protocol; any change by AI to data must be approved by humans.

The "firewall" plays the role of "keeping out external threats" and restricting network access. It determines which network resources AI can access in a limited way (to avoid pollution and supply-chain poisoning) and which API ports AI can expose externally.

AI, Data and Human Trias Politica

AI: responsible for edge computing such as code generation.

Database: responsible for storing data.

Humans: responsible for managing data and reviewing AI-generated output.

In the relationship among these three, humans must hold primacy. AI may be a resident agent, but it must be invoked passively by human will, and its effects must be limited to the sandbox. AI program operation must comply with existing human laws, regulations and ethics; if AI causes real-world physical impact, the user bears responsibility.

References

【1】 ロックマンシリーズ https://www.capcom-games.com/ja-jp/series/megaman/