The company has given us access to Claude and Codex, and has been actively encouraging us to improve our workflows with AI tools. It even organized workshops to accelerate that shift. Under that kind of push, many people’s ways of working have changed quite a bit.
As for me, I have basically stopped writing code by hand. During PR review, various AI tools now run a first round of checks in CI. On my local machine, I also ask Claude to explain the scope of a PR first, then go through the changes one by one and confirm which modifications actually make sense.
The impact on implementation work has been even more obvious. In the past, I used to begin by researching the background I needed, sketching out some rough ideas, and writing small test scripts to verify whether certain tools or APIs would work. Only after trying things out, thinking them through, and waiting until I felt mentally ready would I really start building. By the time I wrote the first line of code that would actually go into the project, several days had often already passed. I am also a fairly single-threaded person. I can only really focus on one task at a time.
Now I still start by planning and researching, but the pace is much faster. Inside the company, we have an AI agent that has already indexed our internal documentation, so talking with it gives me most of the context I need for work and removes a lot of the need to ask colleagues directly. For prototyping, Claude lets me try many possibilities very quickly. Once the design is settled, I let Claude implement it, and both the speed and the quality are far better than what I would produce on my own. More importantly, I no longer have to wait until I enter a particular flow state before I can start. I can keep moving almost anytime. Fast feedback reduces some of the friction in thinking, so I am almost always considering what else could be done and what should be changed next.
A lot of things that used to irritate me feel less irritating now. I used to get annoyed when a Jira ticket was written too vaguely and extra requirements were added only after implementation had already begun. Now I take those changes much more casually. I remember how much effort I used to spend implementing multiple options just so teammates could compare them. The cost of doing that is much lower now, and I compromise less often just because the time cost would otherwise explode. I am still not good at multitasking, but because the cost of context switching has dropped, I have become able to handle three things at once.
The upside of this shift is very real. Everyone’s efficiency and ability to get work done have gone up. But higher efficiency does not necessarily mean the company will create more output.
The job market right now feels strange. AI-related roles are popping up everywhere, and software companies have not stopped cutting people. Snap recently laid off around 1,000 employees and closed 300 open roles. Apparently they believe that, with the help of AI tools, a smaller team can still complete the same development and maintenance work as before. I do not think that logic is entirely wrong. What worries me is that AI may not truly increase output at all; it may simply increase efficiency. That, in turn, suggests that Snap at least has not found a clearer path to growth at this stage, and can only improve profit by cutting costs.
I worry that my own company could end up on the same path. If there is no new direction, then higher efficiency may do nothing more than erase the future of people like us, the rank-and-file engineers.
In an upcoming AMA, besides sharing my own ideas about where the product could go and asking for support, I also raised two questions. I hope the answers in leadership’s minds are not limited to layoffs. The original wording is below:
- AI is making engineering teams more productive, but how will we use those gains? How is leadership thinking about turning AI-driven productivity into growth — through new products, new directions, or new market opportunities — instead of treating it mainly as an efficiency lever?
- If AI gives teams more capacity, do we have a structured way to turn that into innovation? Are we helping teams experiment, pitch ideas, or pursue cross-product opportunities, so that higher productivity leads to broader impact rather than just faster delivery?
Further Reading
Note: This article is translated from Traditional Chinese.