Thoughts on AI Writing

This might be a fairly common sentiment nowadays, but I am getting increasingly worried about the over-reliance on AI writing (and frankly tired of reading it).

I don’t blame anyone for using it though. I use it extensively, but I am trying to be more conscious and careful with how I wield this tool. I think it can dull your mind, and replace your voice. But I’ll focus on the second point today.

In the last couple of years, work as a software engineer has changed drastically. The expectations are higher than before, and agents are really good at writing code now. Obviously they are not perfect, but they still do an incredible job.

As the drivers we should be putting in place proper guardrails and providing the right skills. For finishing touches I often still have an IDE for manual adjustments (I have great respect for coworkers who are able to use agents for the final touches).

We didn’t stop at coding. We aggregate and summarize emails, Slack threads, or any other stream of input. This saves time so that we don’t have to manually sift through all of the information. We’re being encouraged to use AI in all of our workflows to squeeze a bit more efficiency out of our day. I think it’s actually really cool if you can creatively pipe together tools to enhance your workflows.

However, one thing that caught my eye was the encouragement of LLMs for filling out our “Low Level Design” documents, where we spec out the implementation details of a new feature or system.

At first glance, I thought it was fair. It’s a pretty long document to write by hand, so the extra help is appreciated.

I’ve written a few of these documents myself, and most recently I got an agent to write nearly all of it (with manual review and edits after). Then the relevant stakeholders reviewed the document, I got feedback, the design got approved and we finished implementation.

All is great! Hours of my day saved. Thank you AI!

A week or so later, I was reading another design document written by my coworker, and I thought it was very well written. It was clear, funny (humor in a design doc?), and most importantly I could see their voice in the writing. It was unmistakably written by this person and carried their unique style.

That experience left me feeling a bit disappointed in myself for choosing to submit a piece of writing that was written by AI and augmented by me, instead of the other way around. It should be written by me, and augmented with AI.

Compounded by the fact that AI is writing our PR descriptions, code comments, etc… I think we’re being bombarded by AI writing from all directions. Many social media posts and comments are also very clearly written by AI — we always see the em dashes 🙂. I should do my part in reducing this where I can.

In conclusion… there is no conclusion. I’m just ranting and dumping my thoughts here. In light of this recent reflection, I’m trying to avoid blanket generation of text by AI. I feel it’s lazy and comes across as inauthentic to my coworkers and a disservice to myself.

You’ll be happy to know that this was written by my two hands 🙌.