AI随想1

AI 成为基础架构之后，skill和 agent还有各种组件都会变成普遍且廉价的东西，harness eng practice会成为工程师的普遍技能。不会的话会掉队，会了在不掉队的人里很普通。
​
​真正有竞争力的，依然是对业务和技术的系统性理解，对关键链路和细节的掌控，对于ai产出进行原创思考和审查（区别于可以明确出固化下来的知识），这个道理之前就是，现在没有变，只是更加明确。

但这样的关键能力已经很难被积累形成了，因为会越来越少的人去读代码，而这种能力恰恰是读代码积累起来的。读代码这件事不因为AI出现就变得无价值，AI不是确定的软件，和人类大脑一样会出错，他不会成为横亘在人类和软件之间的一层，而是和人类一层去编写和使用软件。对服务方式和品质确定性要求高，面对客户的边缘，会更多使用软件。越多依赖AI，就越少形成知识。
​
​大浪淘沙，靠编程技术门槛吃饭的人，靠隐藏上下文吃饭的人，他们的凭借都已经不再是壁垒，AI可以轻易碾压。
​
一人公司是一个很好的故事，但不是所有的场景都是一人公司，过于理想。团队合作和大集团军作战依然是必须的，商业的较量依然存在，这决定了高效的集团军具备更大优势，当然，使用ai的效率成为关键武器。
​
翻译成英文

Once AI becomes infrastructure, skills, agents, and various components will all become commonplace and cheap. Harness engineering practices will become a baseline skill for engineers. Those who don’t pick them up will fall behind; those who do will just be ordinary among the ones who haven’t fallen behind.
What remains truly competitive is still systematic understanding of the business and the technology, command over critical paths and their details, and the ability to do original thinking and review of AI’s output (as distinct from knowledge that can be clearly codified and locked down). This principle held before, and it hasn’t changed now — it’s just become more explicit.
But this kind of core capability is getting harder and harder to accumulate, because fewer and fewer people will read code — and this capability is precisely what’s built up by reading code. Reading code doesn’t lose its value just because AI shows up. AI is not deterministic software; like the human brain, it makes mistakes. It won’t become a layer sitting between humans and software — instead, it writes and uses software alongside humans, on the same layer. Wherever there’s a high bar for certainty in service delivery and quality, or wherever you’re at the customer-facing edge, software will be used more. The more you lean on AI, the less knowledge gets formed.
The tide will sort people out. Those who made a living off the technical barrier of programming, and those who made a living off hoarding context — what they relied on is no longer a moat. AI can steamroll right over it.
The one-person company is a great story, but not every scenario is a one-person company — that view is too idealistic. Teamwork and large-scale “army group” operations are still necessary. Commercial competition still exists, and that means an efficient army group still holds the bigger advantage. And of course, efficiency in using AI becomes the key weapon.