Published on

AI change my work style

  • avatar


  • When GPT-3.5 first came out, I was still writing Java SpringBoot. I thought it was amazing, but it wasn't powerful enough to help me improve my work efficiency because I was mainly doing business development, and my code was strongly related to the business. (However, when ChatGPT first came out, I still registered it by buying a virtual phone number in India website to try it out.)

  • But with the emergence of LLM like Claude and Gemini, and the continuous development and progress of AI-related technologies like AI-Agent and RAG, AI has started to help me improve my work.

  • At the beginning of 2023, I switched from being a Java developer to a Devops engineer. My primary working language was no longer Java, but became more diverse. I needed to write Shell, Yaml, JavaScript, Golang, Python...

  • After working as a Devops for a while, I found that repeatedly writing Shell and Yaml was very draining. It was tedious and boring, so I started trying to use LLM tools to improve my speed in writing Shell and Yaml. At first, I tried Gemini and Claude, and I found it worked well. (Mainly because Gemini and Claude was free at the time and had no plans to charge.)

  • After using it for a while, I became more and more dependent on LLM. I started using LLM to write more complex programs, help me debug, even help me translate English and solve problems unrelated to work.

  • At the beginning of 2024, I paid $144 for a yearly membership on Devv, which further improved my work efficiency and allowed me to use better LLM models to solve my problems.

  • Then, recently, I started using AWS Toolkit and Alibaba Tongyi Lingma to help me program, help me with code completion, write unit tests, etc. After trying these two tools, I'm start thinking about whether to try Microsoft Copilot.

  • So far, I've talked about the advantages of LLM, such as how it helps me write code, debug, test, etc. Now let me talk about what I think are the disadvantages or shortcomings of LLM.

  • First, I'm becoming increasingly dependent on AI. When I get a problem, my first reaction is to ask AI. Only if AI can't solve it do I start thinking independently about how to solve it. This seems to be subtly weakening my ability to think independently. Second, AI doesn't seem to be very good at system architecture design. It does well at solving one small problem after another, but it doesn't seem to do so well for this type of problem. Of course, Maybe I wrote a wrong prompt.

  • Since 2024, semiconductor companies led by #Nvda have seen a collective surge in their stock prices. Tech giants like the Magnificent 7 have also continued to increase their investment in AI, with #TSLA's FSD + RobotTaxi attracting a lot of attention. More and more AI applications are emerging, and competition in the future will become increasingly fierce.

  • In this wave of AI development, some jobs may be replaced in life, and of course, some new jobs may be born because of AI. In short, the productivity of this society is constantly improving. All I can do is keep up with this AI rhythm and experience more interesting things.


  • ChatGPT 问世以来,AI 就开始慢慢地改变了我的工作方式。

  • GPT-3.5 刚出来的时候,我还在写 Java SpringBoot,当时只觉得这东西很厉害,但还没有足够的能力帮助我提升我的工作效率,因为我主要做的是业务开发,代码和业务强相关。(但 ChatGPT 刚出来的时候,我还是通过购买印度的虚拟手机号进行注册尝了鲜)

  • 但随着 ClaudeGemini 等等大模型相继问世,同时 AI AgentRAG 等大模型相关技术不断发展进步,AI 开始为了我的工作提升助力。

  • 在 2023 年初,我从 Java 开发工程师转职到了 Devops 工程师,主要工作语言不再是 Java,而是变得更加多元化了,我需要写 ShellYamlJavaScriptGolangPython...

  • 而在转职为 Devops 一段时间后,我发现反复地写 ShellYaml 非常消耗的我的精力,繁琐且无聊,所以我开始尝试使用一些 AI 工具来提升我写 ShellYaml 的速度,一开始我尝试的是 GeminiClaude,当时发现有不错的效果。(主要是当时Gemini免费且没有付费想法)

  • 在使用一段时间后,我对 LLM 的依赖越来越严重,我开始用 LLM 写更加复杂的程序,帮我 Debug,甚至帮我做英语翻译并解决一些和工作无关的问题。

  • 2024年年初,我在 Devv 上花费 144$ 开通了年度会员,这进一步提升了我的工作效率,使我能够用更优秀的模型来解决我的问题。

  • 然后,最近我开始使用 AWS ToolkitAlibaba Tongyi Lingma 来辅助我编程,帮助我做代码补全,写单元测试等,在尝试了这两个工具后,我已经在考虑是否要尝试一下 Microsoft Copilot 了。

  • 以上说了这么多,讲的都是 AI 的优点,比如它如何帮助我写代码,修 bug,做测试等等,下面来讲讲我认为的AI的缺点或者不足。

  • 首先,我对 AI 越来越依赖了,遇到问题我第一反应就是去问 AI,只有 AI 解决不了才去独立思考怎么解决,这好像在无形中弱化了思考问题的能力;其次,AI 对于系统架构设计方面好像不是那么优秀,它对于一解决一个又一个的小问题上表现不错,但对于这类问题好像表现得不那么好,当然也可能是我的 prompt 不对。

  • 2024年以来,以 #Nvda 为首的半导体企业的股价集体飞升,Magnificent7 等科技大厂也在持续加大对 AI 方向的投入,比如 #TSLAFSD + RobotTaxi 更是吸引了不少人的眼光,越来越多的 AI应用 开始出现,未来的竞争也会来越激烈

  • 在这一轮 AI 发展的浪潮中,生活中也许会有一些工作岗位被取代,当然也可能会有一些新的工作岗位因 AI 而诞生;总之,这个社会的生产力是在不断进步的,我能做的就是跟上这个AI节奏,去体验更多更有趣的事情。