As a company owner, I am personally experiencing the AI revolution on the front line. In addition to buying Claude Max for everyone to use, our current operations meetings will set aside a period of time for everyone to share their recent experiences and bottlenecks in using AI to assist with work. I use the PM the most aggressively and give the most feedback. Fortunately, my colleagues are willing to learn, and I believe we will soon bridge the cognitive gap among team members. People who are using coding agents seriously must have a feeling. The agent from a year ago and the agent from today are almost two different species. This year, coding agents have gone from being "usable" to "necessary", and their adoption rate is still accelerating. I used to think about writing it myself, but now my first reaction is to leave it to the agent first. Why is the programming field running so fast? Because the code itself is a highly structured and logically rigorous language. There are massive StackOverflow Q&A and GitHub open source projects online, all of which are publicly available, standardized, and verifiable training materials. The conditions for AI to grow by eating these things are really great. But this is not the case in other fields. The details of a case handled by a lawyer, the accumulated clinical judgment of a doctor, and the valuation model constructed by an investment bank analyst have long been locked within the company, not systematically digitized, let alone publicly disclosed. Without sufficient nutrients, AI naturally cannot grow muscles. Silicon Valley has already understood this matter, so in recent years there has been a clear trend: AI companies are crazily acquiring "domain knowledge companies". Harvey AI, a legal AI company, has a valuation of $8 billion after three years of establishment, serving over half of the top 100 law firms in the United States, with annual recurring revenue exceeding $150 million. At the same time, Filevine acquired Pincities and Parrot consecutively, opening a new office next to the Anthropic headquarters in San Francisco to directly compete for AI talent. The same logic applies to the medical field. Stryker acquires Care.ai to build smart wards, and Commure buys Augmedix's AI medical transcription technology for $139 million. Abridge focuses on turning medical conversations into medical records and has raised $550 million in a single transaction by 2025. These cases are all about a fact: capital is using mergers and acquisitions to directly buy the experience curve accumulated by humans over the past decade. Whoever masters domain knowledge can train truly useful vertical AI. What does this mean for white-collar workers? This means that the time difference is rapidly shortening. Previously, AI companies lacked domain data, so the progress of AI in vertical industries was slow. Now this gap is being filled, and the moat in fields such as law, healthcare, finance, and accounting, which were previously considered to require highly professional judgment, is being eroded layer by layer. The replacement of entry-level positions by AI has become a current trend. Six months ago, I thought this would be three to five years from now, but looking at this pace, it's not surprising that there will be structural changes within two years. There will be fewer new job vacancies, and more and more companies will not fill them or even directly lay off employees. People who are not good at using AI will stand at the forefront of a tsunami. Many white-collar workers can only grasp one thing in the end: humans have licenses, AI does not. Just like a proxy and notary, the workflow is highly mechanical, but the law requires a signature from a person to take effect. In the future, lawyers, accountants, and physicians may also move towards this form, where AI completes most of the substantive work and humans are responsible for the final review and accountability. Autonomous driving follows the same path, with mature technology and the real obstacle always being the responsibility. But with accident rates lower than those of humans, efficiency and cost advantages being quantified, many states in the United States have now begun allowing commercial operation of autonomous driving, shifting responsibility to operators and manufacturers. AI replacing white-collar workers will eventually lead to the same outcome. Elon Musk recently mentioned in an interview that Optimus' ability to perform surgical procedures will surpass that of top doctors within three years. It takes more than ten years to train a cardiac surgeon. After learning one machine, 100000 machines can be updated synchronously without fatigue, shaking, or decreased judgment after consecutive surgeries. It takes humans ten years to train an expert, and AI only needs one training session to replicate an entire population. --- As an ordinary person, what can you do? Firstly, to learn how to use AI, you need to be more familiar with it than others. It's not just about copying and posting to chat with Chat bots, but about learning the latest knowledge. For example, trying out Claude Code's latest Cowork release and actively exploring tools outside of ChatGPT. Secondly, don't just focus on the execution layer, be someone who "assists the team in importing AI". Being able to use AI is fundamental, being able to teach others how to use it and help companies import it is the moat. Pure executors are the easiest to be replaced, while those who drive change will always have a place. Thirdly, plan B outside of finding a job. Take advantage of the surplus and start laying out sources of income outside of work. Side jobs, investments, and secondary expertise are all good, don't put all your eggs in the same basket.