As we rely more and more on all kinds of AI tools, I start to believe our work will be more about giving instructions, than actually do the things.
I’ve always been a PHP developer and sucked at JavaScript. Not to mention with the paradigm shifted from jQuery to ES6 and React, we need to write more and more JS code even as a backend developer.
The way I overcome this is to writing all comments in my IDE first and then let GitHub Copilot fulfill the code, at lease the bare minimum of it. The code won’t be perfect and sometimes it can be wrong, but it already saves me tons of time surfing on “can I use” or “you don’t need jQuery” for some very basic syntax or finding the right function to use.
Just the other day, I asked ChatGPT to optimize some JS code and it rewrote part of the code with `.reduce()`. So I further asked it to explain how `.reduce()` works. That’s probably the first time I really understood it.
Good prompt is the foundation of success
Last year when I took some prompt engineering courses, they all suggested you do some role playing or fill up AI tools with the back story before giving task instructions, to increase the success rate.
I didn’t really pay attention to role playing as I’ve added my background in the “custom instructions” of ChatGPT so I rarely need to repeat that. But I noticed both ChatGPT and Gemini nowadays all “self-explains” the whole thought process before they throw the final answers at you.
It seems that AI tools are evolving so quick that the prompt engineering techniques won’t last for more than a quarter, before we need to update them.
The bottom line is, just giving them the instructions as you will give to your human colleagues, in their best interests, to help them understand the whole pictures before coming up with solutions.
Appreciate good instructions
I can’t thank enough when someone gives me good instructions, that are so on point, informative and significantly reduce my time to get into the nitty-gritty part of an issue.
Yet I’ve also found it’s a very rare skill. Most of the time, people do give instructions, but the value from them can be very low.
This is especially exhausting when handling support tickets in software businesses. People often write in to request supports but don’t do a good job at explaining the issue. Sometimes they send in screen casting videos and thought it would help. Except the video is way too long and hard to identify what to notice.
HelpScout is now offering a feature to use AI to summarize a ticket, which can be very helpful. But still, the best would be giving good instructions in the first place. May it be a Pull Request, a support email, or a Trello card comment.
Take it into consideration before collaborating
Personally, I take the person’s ability to give good instructions into consideration, when deciding if I would take them as a client, or hire them for a project.
Indeed, if they give just regular instructions, that doesn’t mean they are unqualified. But if they give good instructions, it’s like 200% of the chance, they’ll be the best to work with.
I apply this rule to myself too, of course. I always try my best to write good instructions, sometimes I do think too much time is spent, but hopefully it worths. It applies to almost all written materials, as I work remotely with all my clients, I usually communicate with written messages.
One last tip is do not ask open questions. Be precise about what you’re looking for (after giving the instructions). Even add the next action step so the others won’t be left in wonders. Mainly, in the sense to save their time other than save yours, is a good rule of thumb.
Leave a Reply