I spend most of my weekday writing and reviewing code, and for the past year, a huge portion of the workflow involves a digital infantry of AI bots:

• Cursor
• Claude
• GitHub Copilot
• Atlassian Rovo
• Random Slack integrations

My feelings about these tools honestly change day-to-day, and they are certainly useful and boost “productivity”, or at least code output. This also comes at a cost incurred by my company that has been rapidly increasing each month. I’m not going to focus on these tools today, though.

On Writing (or Not Writing)

¹

A few months ago I came across this post by Bryan Cantrill by way of Sam Firke’s blog. Since I read the term, I have thought about one’s intellectual fly being down daily, and have sent this article to a handful of colleagues, and it is something that most people seem to resonate with. I tend to feel a kind of embarrassment and unwillingness to read something that is clearly not written or even understood by the author.

I am not optimistic about AI-generated writing, and think that most people should not use it when communicating with other humans.

Distilling some of the fundamental problems that I’ve seen and read about:

• Before AI tools could produce realistic sounding articles and documentation, the mental burden and ownership of information was purely on the author.
• Now, the burden of assessing the validity of a document, whether it contains accurate information, and if you should even spend your time questioning the details is entirely on the reader.
• AI-generated pull request summaries tend to have accurate and sometimes useful factual information, but often miss the mark when communicating the intent and impact of changes.
• Even with strict prompting and IDE setup (using skills, rules, etc.), LLMs tend to add overly-verbose comments which can sometimes be wrong, misleading, or both.

Communication is such an important part in building software, technology, and especially while considering new approaches to problem solving. Even in less technical roles, I strongly feel that I was able to learn a ton about the people I work with every day by reading their writing in messages, emails, and of course during face-to-face discussions.

For some people, typos can indicate that they’re rushed, exclamations show that they’re excited or angry, and punctuation typed by real human hands tends to align with how the person typing is actually feeling. All of this seems really obfuscated now, and for everyone’s sake, I hope that we can settle on a more authentic way of communicating in the workplace, online, and anywhere else.

Footnotes

1. I made this in Affinity! Clearly not an artist, but I like to doodle. ↩