AI and the Future

At this stage, when every company is using AI to improve efficiency and cut staff, I wanted to write down some thoughts on how AI may develop from here. But as soon as I sat down, I instinctively opened chatgpt and wanted to look up a summary of the recent completion era, the vibe coding era, and the harness era that is now actually becoming common. That was when I realized AI has already been woven into life, and that I have not done much independent thinking for quite a while.

Looking back carefully, it took less than a year to go from the completion era to the start of vibe coding, and less than another year for vibe coding to become basically usable. The changes in between have deeply reshaped the development paradigm. And now, in order to better make large models fit the user's development standards and needs, people are using harness engineering to control them.

It is not hard to see that the way coding evolves has gone from model prediction, to model execution, to better model execution. The final harness paradigm is really about constraining the GENERAL capability of large models. And the way harness is expressed has also moved from code rules like lint and prettier to natural-language constraints written in md. That is because, at this point, the gap between natural language and code between models and humans has already been removed. And this development style will continue to exist and keep evolving before the stage where entire repo and project automation becomes real. Of course, some companies have already started exploring fully AI-automated execution. For example, jules and stripe's minions.

Here I want to make a few predictions about future AI trends in code development. As for why I am choosing code as the angle, it is because most of the digital software, hardware, and their derivatives that humans build with code are still the fastest and broadest way to affect real life.

Code Management and Development:

We are still in a chaotic period. Company-side AI efficiency pushes are causing layoffs, and laid-off employees will flow into traditional companies to drive AI-enabled transformation, or start businesses, or build personalized products. That may create a short-lived code boom. As AI becomes a longer-term part of daily life, most people will gradually master the related skills, and in the end the code for everyday applications in the same domain may converge into one. For example, perhaps only one IM app remains, like telegram, while more specialized software in niche professional fields still keeps its own differences. Users will only need to raise an issue, and the applications used by the general public will be implemented and maintained by AI. Different needs will be handled by different plugins, and at install time AI will automatically choose and insert them according to the requirement. That way, the app can keep a consistent core while still satisfying different needs. Most token usage will then be spent on harder problems such as mathematics and physics, migration to other planets, environmental balance, and other major challenges.

Deployment Model:

In the future, AI capability will be too general-purpose, so it will likely be split into different capability layers. The most commonly used natural-language intent handling will be separated out, and the code-related capability will also be split out. The two will then be combined for ordinary development work. The most public and open part, natural-language intent, is more general and has much higher traffic, so that part may be deployed locally by companies, while the provider's own private model code capability is deployed in the cloud. That would help control cost and improve performance.

Further out in time, we will also have to think about the influence of social ideology. I hope the future is one where everyone is equal, everyone can afford to eat, and we do not return to a feudal society again. Probably no one will ever read this post. It is only a small marker for myself. I just hope that if I am still alive and read it again in the future, I will remember these anxious days. These are only my shallow and rough personal thoughts. That's all.