Happy weekend,
In case you haven’t seen, Northeastern has enabled Claude Code on our student accounts! If you have little experience with coding agents, it’s a great opportunity to dip your toes in, and might radically change how you view the problem of writing software (hah).
Remember that even the frontier LLMs have poor taste relative to a human. In a domain you are unfamiliar with, they can act as either a teacher, a crutch, or both, depending on how you use them. These tools are incredibly valuable when you already know how you want the code to look, but want to save on the mechanical labor of typing it out. They also function as surprisingly good code-reviewers.
I’d caution against “results driven” prompting, wherein you do not think through the programming portion of the problem, but exclusively the client-facing result of solving the problem, for a few reasons. First: everyone has access to these tools, so unless you are actively applying your expertise, you are genuinely on equal footing with a random middle schooler. Secondly: the code quality produced this way is very low. Lastly, you will genuinely shed your intelligence.
As briefly mentioned, you are free to use coding agents on your tickets, however you are still entirely responsible for your own committed code. Always review agents’ work thoroughly, and push back when you would personally do things differently, even if it’s small. If you do not understand the outputs, seek to understand long before ever committing.
Have a nice weekend everyone!