Which one are you?
Three paths for engineers in the age of AI
Different tools are superficial, workflow is the watershed
Recently, I was talking to a friend about the way they write programs with AI. The differences were outrageous - not in tools, but in workflows.
There are three broad categories.
The AI catches bugs, predicts problems, and occasionally patches. Engineers decide what to write and how to write it, AI doesn't touch the main process.
The development is left to the AI, and the engineers specialize in writing tests. Define "what's right" and run it once to know if the AI's code is right. Work has changed from writing programs to writing specifications.
Development and testing are all handed over to AI, describing requirements, reviewing output, and adjusting details. Don't touch the code, only the direction.
When ChatGPT 3.5 came out, I subscribed to ChatGPT Plus to write my own code, and once in a while I dropped a paragraph to ask GPT if there is a better way to write it. Later on, my company bought Team program, and my personal program was returned.
Started using Claude Pro, bill jumped to $200 a month, AI wrote more codes, spent more time testing and reviewing myself.
GitHub Copilot, for its workspace functionality and GitHub integration. It's one of the tools that I've been comfortable using in the VS Code plugin for a long time.
Google AI Pro, which came with NotebookLM and Nano. It's free to use for U.S. schools.
While playing OpenClaw, I accidentally learned the skill of multitasking. The models were smart enough that I was able to do things with AI that I never thought I could do before - OpenClaw was too token intensive and was shelved - but that experience turned me into a total commander and I turned to making my own stuff.
| service | Period | Amount |
|---|---|---|
| ChatGPT Plus | 2023/2 ~ 2024/2 | $260 |
| Claude Pro / Max | 2025/2 ~ 2026/3 | $1,342.86 |
| GitHub Copilot | 2025/5 | $100 |
| Google AI Pro | ~2025 Mid ~ Present | $0 (edu free) |
| Total | $1,702.86 | |
Commandos don't require you to be an engineer. Anyone can tell an AI to do something.
But the difference is enormous.
For those who don't understand technology, AI will take whatever it gives. I can't tell if the architecture is good or not. I don't know where the potholes are.
People who know technologyWe can draw the boundary first - how to choose the frame, how to dismantle the modules, and where not to move. AI works within the frame, and pulls it back if it is crooked.
The difference is not in the speed of production.At the time of the accident.The
It's like a software department. The director knows the technology, but doesn't usually write code, sets the direction, divides the tasks, and reviews the output. The engineers at the bottom do the actual work. Usually it's smooth, and the supervisor only has one mouth to feed. When something goes wrong, the supervisor jumps in and takes over. The supervisor jumps down to take over. Because I know, so I can save.
Output is stable but slow, and as AI capabilities continue to improve, the efficiency gap between sticking to handwriting will only widen.
The leverage is high, but so is the testing capability. If the testing is not good enough, the AI will not be able to test what it delivers.
It's the fastest and easiest, but it's also the riskiest. Your value is one thing: whether you can catch it when it happens.
The three modes are not mutually exclusive. Many people look at the nature of the project switching.
But one thing's for sure: no matter which way you go, "tech savvy" doesn't matter any less - it matters more.
Because the more powerful the AI is.The ability to tell if it's right or wrong.The more valuable it is.
I'm still learning. Commanding is commanding, but LLM brings things to me every day that I didn't expect. Sometimes a prompt comes back with a better result than what I had planned.
The AI community is extremely active right now. New tools, new frameworks, and new ways of playing are popping up every week. While absorbing new knowledge, you have to constantly verify the feasibility - what works, what's a gimmick, and what's worth diving into.
Both breadth and depth are expanding at the same time. There is still a long way to go.