The job market is challenging in the current day in age given that teams can become very productive using AI to enhance the workflow. In the last week, I've updated the following GitHub packages using AI. I didn't just give a prompt to an AI and called that on its own good - rather I spent hours on each utility going through performing a code review and security audit of the packages and then providing recommendations, suggestions, and considerations on how to resolve and/or mitigate the risk. The result is over the last 170 hours I've updated the following packages!
Bump
This package had a handful of bugs that weren't caught by the original test coverage. By asking AI to write more tests that covered different scenarios than were already documented, I uncovered a handful of bugs and was able to squash them. A new type of test was added to main_test.go that is connected to make test-integration that is connected to the existing unit, fuzz and benchmark tests. Both this package and igo have a test.sh script that extends beyond the three (3) types of Go testing. When this fourth type is added, it catches bugs that show up in the wild but that traditional testing misses.
https://github.com/andreimerlescu/bump
Verbose
This package received a formal license for open source use (Apache 2.0), added tests and enhanced efficiency around function calls. It is a novel Go package in that it allows you to effectively build a pipeline application that will knowingly handle sensitive information and the verbose package allows you to register that which should be protected and expose that which should be exposed to STDOUT versus filesystem protected files. verbose will censor to STDOUT and write values to file.
https://github.com/andreimerlescu/verbose
Goenv
This package received minimal updates this month. I removed from the Makefile the summary task associated on each test and then expanded the README.md documentation further.
https://github.com/andreimerlescu/goenv
Entropy Password Generator entpassgen
This package did well in its review and only received a handful of recommendations that would add stability around the loading spinner, and dozens of new tests - all passing naturally.
https://github.com/andreimerlescu/entpassgen
Lemmings
This package is brand new as of last week, so I built it in less than a week with AI. This is because I needed this utility in 2015 when I launched Trakify and was on Barron's Magazine. Shockingly, it never existed on the market until now. Apparently the work involved in building it using OOP was simply not worth it but in Go? Give me a few prompts and we're golden.
https://github.com/andreimerlescu/lemmings
Generate Word Password genwordpass
This package got updated by adding additional words, improving performance of core functionality, and updating the README.md for the repository.
https://github.com/andreimerlescu/genwordpass
Room
Also this week, I created the room package that is capable of giving you a waiting room type of functionality in front of a high traffic website. In the event that a spike in traffic occurs, you don't actually need to have scaling and such in order to stay online - you can use room.
https://github.com/andreimerlescu/room
Sema[phore]
This package got updated so that Try* funcs could be introduced to the semaphore primitive. I added additional tests and functionality that make this semaphore package my go to.
https://github.com/andreimerlescu/sema
Conclusion
AI has made the volume of jobs on the market dwindle, but AI has increased my productivity by at least 10x per week. I am able to accomplish this because I do not require AI to get me from zero to hero. I only need AI to help me scaffold out quick rough code and then I can patch it over and fix it up and then hand back the next step for the AI to work on. It's iterative. The AI doesn't know how to build what I am building in a one shot - but I understand how to build with AI in a much faster rate than I ever had in the past.
Given this, I feel like calling myself a software engineer is pointless. I build solutions to problems, and I use software to accomplish my goals. I need a new title to call myself. The strange thing is, in my professional career, I've cared very little about what my actual title was.
If I was given a $100K AI budget per month, I could use it wisely to actually accomplish really hard problems. Instead, I work off from a $20/month Pro Claude subscription and leverage minimax local LLM when I am in cool down on Claude. That happens every day after 30 minutes of usage. That's what $20/month gives somebody like me. And for that, I can build this with it in a week.
Top comments (0)