Overview
Friends who have been following my blog may have noticed that over the past year, or more accurately the past two years, my posting frequency has been extremely low, with very little high-quality output. Over the last year in particular, I basically stopped updating altogether. For someone like me, who has long been used to thinking and summarizing through writing, this kind of prolonged silence usually signals either intense turbulence in my inner world or a deep restructuring of the external environment around me.
That is indeed what happened. Over the past two years, the most obvious change was that I moved to ByteDance, where I had to adapt to the ByteDance style and take on greater challenges at work. At the same time, the rapid development of AI once made me feel that blogging had become somewhat unnecessary. If I wanted to learn something, I could just ask AI directly. And for hands-on tasks, I could have it guide me step by step. But after adjusting my mindset over the past few months, I’ve realized that although AI is now very powerful, maintaining independent thinking has become even more important in the age of AI. So I decided to pick blogging back up again.
Career Changes
After joining ByteDance, the role I took on changed in a fundamental way. In the past, my work was more focused on diving deep into a single module. At ByteDance, however, I’ve had to deal with the more demanding “ByteDance style”: not only being a core developer, but also taking on broader ownership. That means helping define the technical direction, planning the evolution path, and coordinating with team members to break big goals down into concrete execution steps.
The most profound challenge has been cross-business collaboration. In the ByteDance context, building a feature or delivering a project is only the beginning. The real test is how to align multiple business lines, convince them to integrate your solution into real business scenarios, and actually use it. This shift from “building behind closed doors” to “selling and enabling” requires technical vision, communication resilience, and a precise understanding of business pain points. Those were dimensions largely missing from my previous professional perspective.
The changes in the kinds of projects I’ve been involved in since switching jobs have also given me a more concrete sense of how large the gap in technical evolution can be.
I still remember that three years ago at Shopee, a colleague once shared a panoramic roadmap for Service Mesh. It was a grand blueprint covering foundational components, platform construction, traffic governance, observability, and more. At the time, it felt more like an idealistic vision than something real. But after coming to ByteDance and working in the current business environment, I was surprised to find that many of the ideas in that roadmap, which had once seemed two or three years away, are now part of my daily work, both in terms of usage and development, and have already matured into industrial-grade solutions. That transition from “vision” to “infrastructure” gave me a deeper understanding: designing an impressive architecture is certainly admirable, but finding the right insertion point to actually land that architecture is where an architect’s real ability is tested.
The Development of AI
Over the past year, the AI field, especially around large language models, has been evolving at a breathtaking pace. The pressure comes not only from rapidly changing technical benchmarks, but also from company-level expectations around AI enablement and dramatic productivity gains. For a time, I found myself repeatedly questioning my own competitiveness. That fear of “falling behind the times” was also one of the internal reasons why I stopped writing on my blog.
Through actually using AI and mentoring junior engineers, I came to an interesting realization: to some extent, AI really is replacing part of the responsibilities traditionally handled by junior roles.
As a mentor, when guiding slower-paced newcomers, I often have to describe task requirements in extreme detail, sometimes going through two or three rounds of repeated communication just to make sure the implementation does not drift off target. Strikingly, this feels very similar to working with AI. When the requirements I give are not specific enough, the output from AI can be far from what I expect. Only after several rounds of adjustment and guidance does it reach the standard I want.
That comparison made me realize that AI is, in essence, like a young worker with near-infinite knowledge but an extremely high dependence on clear instructions. As AI continues to improve, entry-level roles that can only passively follow instructions without independent thinking will see their room for survival shrink rapidly. For people whose job is to set direction, the new core competitive advantage becomes the ability to define problems precisely and to “coach” AI the way you would coach a junior engineer.
Once I truly calmed down and started building, debugging, and experimenting with AI applications myself, my anxiety gradually faded through hands-on practice. I realized that although public discussion is full of hype and exaggeration, once you strip away the surface, there really are meaningful breakthroughs emerging at the foundation that can reshape productivity.
The best way to confront the unknown is to become part of it. Rather than standing on the shore endlessly estimating the height of the waves, it is better to jump into the water and feel the current for yourself. When you genuinely participate, define direction, drive implementation, and guide AI the way you guide newcomers, you begin to see that real technical growth always happens in the process of solving problems.
Some Small Plans
So in my upcoming updates, I want to focus much more on AI-related content. This will include not only hands-on reviews of various AI tools and frameworks, but also more practical topics such as how to introduce AI into the full R&D workflow and how to improve team collaboration through prompt engineering.