Thoughts on Vibe Coding

@Creed_Shen|Jan 1, 2026 (1m ago)5 views

After an intense weekend of vibe coding, I am even more convinced that AI will eventually replace humans for writing most code.

AI already knows the syntax, APIs, and best practices of languages and frameworks better than most people. If that is true, there is no reason not to let AI handle most implementation work.

So what is the difference between professional developers and ordinary users doing vibe coding? Why would a company still hire developers instead of just asking anyone to prompt an AI? I think the answer is in a few key areas.

Developers communicate with AI more efficiently. I can no longer say "non-developers cannot read code" as a hard barrier. People can ask AI to explain code in natural language and then request changes. But developers can skip that translation loop and, in many cases, spot issues immediately.

Beyond reading code, developers make better decisions because of engineering judgment and experience. This matters most in architecture and solution design. Developers ask better questions and can take long-term responsibility for systems.

Developers also review. Yes, AI can review too, but AI is not something we can trust 100%. AI can make mistakes, and prompt/model poisoning is a real concern. If we depend on AI too heavily, we may eventually fail to notice when the output does not match our intent. We should not let code become a black box controlled by AI. Not every problem can be validated simply by running the generated output.

By the way, how much of AI output do you actually read? 100%, 50%, or almost none? In company work, maybe you still review carefully. But what about side projects? I used to read almost everything. Later I read less and less, and now I only deep-read when I need to debug. If I skip reading and something breaks, AI can still re-read code and fix it. For personal projects, speed often wins. But what if one day we stop reading code for too long and can no longer understand it? Unlike code, AI does not always produce the same output for the same input.

So I do not think software engineering as a profession will disappear soon. But as vibe coding becomes mainstream, turning ideas into products gets easier, and great ideas become more valuable than raw coding ability.

This is an era of creation. AI dramatically boosts our creative throughput. The hard part is no longer building, but standing out from the flood of AI-generated noise.

I work in a traditional fintech company. We have internally deployed open-source models, but they are usually generations behind (traditional industry inertia plus the rapid pace of model progress). On top of that, internal model APIs have strict context limits far below what the models can actually support. Meanwhile, the company keeps saying it is "all in on AI innovation" and expects everyone to produce AI outcomes.

I do not understand companies that talk about AI adoption but still block developers from fully using AI coding tools. In my view, they have no future and will eventually be replaced. It is not about money. If compliance is the concern, large companies can self-host models properly. But many cannot even provide adequate development resources. The productivity gap from modern models is beyond what old management understands. Most leaders do not write code, and many will never seriously try the latest models. Expecting innovation under these constraints only means losing to companies that embrace AI for real.