The boundaries of AI capabilities are constantly changing.
After 2025, large models will no longer be just chat tools.
Long context, Computer Use, Agentic Coding, and multi-step tool calls are all pushing the model to the "task executor".
This means that it can take over more and more structured, repetitive, and verifiable tasks: writing first drafts, changing codes, organizing data, generating documents, running processes, and doing multi-version adaptation.
But it is not good at truly high-density judgments: defining product direction, understanding the implicit rules of the industry, judging candidate potential, doing complex negotiations, and building brand mentality.
Therefore, when using AI to do things at 10x speed, we will first break it down: what is 80% that can be automated, and what is the 20% that must be maintained by humans.
The boundaries and writing between organizations and AI, the collaboration model and interest redistribution after AI empowerment within the organization, new SOPs, etc.
are all problems, large and small.
This also explains why even in 2026, some companies’ understanding of AI still remains at the tool level, and some companies have blind worship of AI, especially middle and senior management who have high ambitions but low power.
The gap here means that even by 2026, there will be very few people who can calm down and seriously use AI to serve the company and existing products.
You can read this long article I wrote, maybe you will get some inspiration.