Software engineering has become prompt engineering and code reviews | Hacker News Reader

Software engineering has become prompt engineering and code reviews

2 pointsechollama4 comments

i have been an ai engineer for nearly 5 years and have built everything from mobile apps to saas products. i've been using cursor as of lately and i feel almost "out of practice". i code more and have a higher level of output but honestly at this point im not sure how efficent i could be without an LLM. i see no reason to stop using it because the output difference is so great and im building products that need to move fast, although im scared in a year or so ill be too dependent. is anyone else feeling this from coding assistants or is it just me?

View on HN

I will say that by using ChatGPT, Claude, and Copilot I have been able to spin up from basically nothing and little to no understanding of the tech to a FastAPI and SQLModle BE, a React Native FE with working API calls, log in/out authentication and a bunch of app features in 5 weeks. Learning and growing, reading documentation and asking the LLM's for help and input. I again feel super productive but also feel a little like a sham as I did maybe 40% of the actual coding. I feel that the prompt engineering part is true but if the person prompting the LLM can't understand the code then eventually the LLM will break things and the dev will have to fill in. I feel that things will get more like this as time passes. Soon it will be engineers prompting LLM's for the actual code and preforming minor fixes to the code spat out.

Could you share more about how you do it?

Also, in a year the LLMs will be even more powerful.

i usually just have cursor chat open and i code some function, get cursor to do a code review of that, look for errors and do some chain of thought reasoning with o1-mini and its related functions. I then run the tests and put log output through cursor to identify bug fixes. Using LLMs as almost a pair programmer as well as something to check my work as im working.

yeah very similar. chatgpt and esp. copilot are pushing me to new heights of productivity. But in an interview I'm getting progressively uselesser. e.g. in the VSCode python debugger where copilot isn't active, I often don't know what to type. I'll sometimes even move back to the editor and type a comment with what a I want, let Copilot fill it in, then paste it into the debugger.

It's maybe sad, but from a business perspective it's excellent.

Reply on news.ycombinator.com