No AI messages 🤖

please don't use LLMs to talk to people

If you don't have time to write it yourself, then why should we read it?

❌ Don't do this

Riley 9:41 AM
Hi Morgan, I see your feature is still marked In Progress, but I'm pretty sure it's in prod, right? Did you forget to update the ticket?
Morgan 9:44 AM
Hi Riley, I understand your concern — this certainly is cause for investigation. Let's summarize:
  • Feature status: I can confirm the feature was finished yesterday around 4pm, here's the commit.
  • Production release: This morning a job ran which pushed everything currently on main into production, so you're correct that the feature is currently live.
  • Ticket status: After checking the ticket in JIRA, I can see that you're right and it's still (wrongly) marked "In Progress" when it should be "Done".
In short: your hunch was right, the problem was the ticket status, I've gone ahead and changed it to Done, thanks for flagging!
Morgan's reply is quick and professional, but also unnecessarily detailed: it was obviously written by an LLM. Riley now has to parse through a long message to get the answer to a simple question!

There's a few reasons why someone might do this: they might be trying to sound more professional, or maybe they're busy and it's faster to get Claude to answer the question. But this example highlights a few common issues with LLM messages:

These problems hold true for ticket descriptions, announcements, documentation... etc. If it's meant to be read by someone, consider writing it yourself! LLMs are incredibly bad at keeping to the bare minimum information required by your interlocutor because they don't have the social context of the interaction. But you do!

✅ Instead try this

Riley 9:41 AM
Hi Morgan, I see your feature is still marked In Progress, but I'm pretty sure it's in prod, right? Did you forget to update the ticket?
Morgan 10:08 AM
woops, my bad, I marked it as done! Thx
Here Morgan took longer to respond (she's still busy after all) and her message is clearly a little rushed, but the result is far more positive for Riley:

❓ What's wrong with using AI?

Nothing! We're still in this technology's infancy, and we're all figuring out what it can and can't do. LLMs are surprisingly good at a lot of things: they're great at reading large amounts of text, including code ; they're ridiculously good at decompiling binaries ; and skills let people automate tasks in a few minutes where they would once have taken days. This is all great, and while there are many arguments against the wide use of LLMs as a technology, there is no denying the productivity gains can be huge!

BUT, that doesn't mean you can automate away social contact, and it doesn't mean that a time saver for you is a time saver for the people you interact with. Before you automate a task, ask yourself what the intent behind it is, and ask yourself what the experience will be for the people who read the final product. You are good at what you do, and when you use an LLM to do it instead, you are cutting a corner somewhere, and sometimes it's worth cutting corners, but sometimes, it's not!