I do stuff because I want to, not for maximum efficiency.
If something can be done in 10 minutes, then fine but if it’s something 15 programmers do every single day forever then maybe the automated equivalent is worth that work.
Conversely sometimes it isn’t and part of the role of being a senior / principle developer is knowing when something is worth the effort and when it isn’t.
Someone once told me a story about how they were still regularly using code I wrote several years after I had left that company. I may have overcorrected based on that one data point.
Yeah but I do it 100x a day and I spend a lot of time handing doing the task on top of actually doing it.
Pulled up my phone and seriously thought this was a picture of me from like 15 years ago. Then I realized how common the look is among this industry.
Reminds me of a time a couple years ago when I took a brewery tour around Boston. Every. Single. Guy. Including myself and the driver/tourguide…looked exactly like this guy, with a beer gut and a plaid shirt.
If a task takes 10 minutes to do manually that doesn’t mean I’ll start now, it might take me more than 10 days to start - so automating it is the faster way for me to get the task done.
Yeah but next time I have this exact same super specific issue I’m going to fuckin fly man.
In some cases, it is “code as documentation”. Saves me from 1) remembering the steps 2) means I actually doing the steps correctly 3) Actually remembering to do the 10 minute task when I need to.
If I had not done the program, then I often would have to write down and maintain documentation elsewhere.
The modern one:
When there’s a function that can be written manually in 10 minutes but you find a way to do it in 10 days using LLMs
“Here’s my Claude skill that takes the LLM 20 minutes with constant babying that saves a human 1 minute of work”
Promo time baby!
And the steps aren’t reproducible because the LLM does something different every time. So you can’t consistently recreate it.
That is my favorite way to notice my knob coworkers are sending me slop PRs. The same function written 3 different ways and all three have a different bugs.
Even better, now you can automate it with AI, so that it’s both very fragile and dependent on a paid subscription to work.
Yeah but since you didn’t write it you won’t remember you generated it and you’ll just have the LLM spit something else out the next time you need it.
Just use this chart and you’re set
What this chart forgets is that you can share your solution. You share your automation with a team of 20 or post it online and the payoff is much higher
Came here to post this too :)
I love this chart, have used it more than once 😄
But something it doesn’t factor in is when you want to automate something to remove the human error on a repeated task. Having to correct a report you already sent to the customer is not fun
Of course there’s an XKCD for everything, thank you
If it’s a task that has to be done every day for 4 years, you’re actually saving time. And that’s if we assume only one person has to do it, if it’s 10 people, you’re saving time after 5-6 months. 10 minutes adds up quickly once it starts scaling.
It’s more than that for me. I typically hate doing repetitive tasks but I love automating them and seeing my scripts or apps run or just checking a dashboard feels better.
Assuming you set it up perfectly the first time, instead of it breaking for some reason six weeks later and then you spend three hours figuring out/remembering how you set it up in the first place and another two nights after work fixing it so it works as intended.
Also your use case has changed, so it needs to be retooled to address the new situation.
Other than that… worth it.
5 years is also a pretty arbitrary span to go with. You could smoothly discount future time savings instead, but then your discounting curve is arbitrary.
The most rigorous way to go would be to set some kind of future goal, and then work your way backwards to find some kind of statistical description of the shortest path there, or some kind of future metric at a specific time and the path that maximises it. This is pretty much how you design your investing portfolio, just with money instead of labour.
There’s also a relevant xkcd for that.
spend three hours figuring out/remembering how you set it up in the first place
I mean, you could also leave comments in your code, but that’s just me
My comments:
// Do not change this timeout value. There’s a compatibility issue that causes requests to hang. Come back later to add details
// Should probably refactor this bit
// drunk, fix later
Just one layer removed from the problem.
“WTF does this comment even mean?”
