Diagnosed with “Being a 10x engineer”
— the curse and blessing of becoming one
tl;dr: Chatgpt 5.2 and gemini 3.0 has both diagnosed me with being a 10x engineer. After a long discussion, short conclusion: 10x is an engineer with wisdom, not heroics, speed, or individual output.
I prevent organization making early wrongs direction that cost future work to reverse
the bless:
Case1
White labelling impact: one of the employer I worked for decided to add white labelling feature. My immediate response was that borrow from OEM in manufacturing, you lose brand identity and profit margin. I asked if it is the intended goal. The high awareness beyond engineering feature is a rare trait when compared to engineers who only focus on ticket completion.
Case 2
I have already been said to have overengineer a take home test where I am told to process a spreadsheet like work of postfix strings. I learnt basic rust and implemented non idiomatic rust with type system to prevent raw token to go to result token directly. This result is that too many Tokens, each enum for each step is just not acceptable to the code reviewer. A data engineering data flow model approach to a take home, no matter how easy the idea is, seems not accepted, expecially later on I found no one on linkedin in the company has ever had any data engineering experience. This is the pain of a 10x/staff+ engineer.
Case 3
When given a vague requirement, I forumulate with iteration of core feature that must not be wrong and focus quality on those, while allowing API for system boundaries. After some time an end to end system is working with error/ hallucinatino checks/ mitigation/ corrections / prevention. Structure preserving is what often neglected. With given readme and operation manual for junior developer to operate. The junior had tried to vibe-code implement own error prevention by wrapping the whole prompt, reasoning, document presented to LLM from what I designed with just another llm run and ask to check. But for many who know it well, it does not work well from cost effectiveness point of view. If the checking LLM is smarter, you should just use that smart LLM to run it in first place. If that LLM is weaker/ dumber, how can you entrust it with a weaker LLM?
Get Hung Cheung Chan’s stories in your inbox
Join Medium for free to get updates from this writer.
There are too many innovative way to get a system new redundant parts and exhausting it with documentation can never work, it has virtually infinite dimensions to get wrong.
Another futile approach is when not understanding deep enough how the system instruction work in gemini cache caused an error. The junior replacement tried to cache an llm conversation with a different output schema, which does not work as the cache hit depend on the hash of the system instruction where the output schema lies in. This is a lesson to those start up who try to get senior staff + to build a working system with core harden but not yet complete and think can cut cost by dumping that senior with the junior hire. Sometimes it hurts when you see what you built to be downgraded to a non-elegant system.
It is just the same as a system, if it is senior implemented designed, how can you entrust the update from a junior developer with the senior dumped? This is wisdom from example.
This is where 10x engineer with architect instinct help company to prevent cost wasting approach. The beauty of system understanding and less-is-more.
Case 4
Pytest clean test flow deleted as no junior developer knows pytest well enough.
It is not I made super clever stuff that noone knows, but just the developer do not know pytest and decided to use QA to manually test.
I made a workflow, spin up a pgsql database container, preseed it as a conftest. Run CRUD workflow (yes, staff+ also need to do CRUD shop stuff).The end result is, the who test flow get deleted, and I was said to be vibe-coding too much. The real reason is the test case was not run properly with proper config, the test that should not be automatic was run by CI developer as automatic and fail to pass it when the documentation already said the test case should be run with exclude manual cases. Too hilarious. And the test is not just disabled but just totally removed from code base.
Maintenability is in the eyes of beholder. If the maintainers are too weak (sub standard to be honest) , you cannot help.
Back to the story, How did I found out that I become one?
When back in high school, I was chosen to a selection team to represent school for ACME/ TOP coder type of competition and I have been a second tier. The top tier has been in google, amazon, yahoo research, university professors. As a second tier to them, I never thought I become one. It has been for many years until I applied for a recruiter called tenas talent and I never care until I found it sound like 10x talents. It woke up my memories back in high school those top tier had once mentioned 10x engineer (yes, long but well remembered memory). And if I was chosen by recruitor for interview, am I one of the 10x engineers? I have this question in my head. I recently asked the status of my previous pair programming trainee of. where they go. One is now in Meta and another in IBM (I even couched the Canva junior ML engineer about Exception and BaseException during the python interview, yes, they interviewed senior with juniors and test case are bugged). I seemed not too bad. I finally took the courage to ask. And thanks to the AI boom, Chatgpt had many of my design conversations and Gemini can check. I bluntly ask chatgpt if I am a 10x engineer. chatgpt give me a positive. I realize that even I was second tier, I am a late boomer to become one now. I then cross validated with Gemini and network definitions. Less code is more, architectural literacy. abstraction good for real projects but bad for take-home. Fractionaly CTO functioning in start ups. Only found to jump in to solve hard problems. I actually had become one of the 10x without knowledge, gradually over the years.
