mikasisiki

1. In the same amount of time, AI now generates about 5× more code than before.

2. In the same amount of time, engineers have to review about 5× more code than before.

3. The time spent reviewing code is now greater than the time spent writing it.

Conclusion: In many companies, AI has actually reduced the overall efficiency of engineering teams.

If companies really want AI to improve productivity in production environments, the key is to improve the quality of generated code and the efficiency of the human-in-the-loop validation process.

What teams actually need to find is a balance point, and that balance point depends heavily on the methods used by individuals and teams.

At that point, AI might generate only 2× more code instead of 5×, but the time required for both individual and team code reviews could drop significantly. In that case, code quality would not decrease—in fact, it could improve—while iteration speed would still increase in a meaningful way.

“Painting” has been a key vehicle for transmitting Western civilization, whereas in China this role has been taken over by “poetry.”

Painting is like a program, while poetry is more like a continually evolving set of requirements that must be aligned with over time.

Perhaps world models are meant for writing programs, while large language models are meant for describing the world. Perhaps we shouldn’t make AI work for humans at all, but instead give it a space where it can create and explore freely.

Most companies don’t just need standardized SaaS tools — they need customized solutions like tailored CRMs, ERPs, Retool-style internal tools, Notion-like platforms, or Salesforce-level systems that align with their unique, self-driven business logic for data creation, modification, and management.

How can AI help build and adapt these personalized systems more efficiently than traditional software development?

Does this demand really exist?