Harry McCracken

Two reasons why I’m not interested in letting AI ghostwrite for me:

1) Writing is how I think things through and come to conclusions.

2) Above all, I am motivated by the possibility that I have not yet written the best thing I’ll ever write.

https://www.wired.com/story/tech-reporters-using-ai-write-edit-stories/

Meet the Tech Reporters Using AI to Help Write and Edit Their Stories

Independent writers are using AI agents all throughout their reporting process. What’s the value of a human journalist, anyway?

WIRED
Mar 27 at 3:30amFedica
Show threadZazzoo 🇨🇦2d ago
@harrymccracken Yes, but it's also true you haven't yet written the best thing somebody else already wrote.
Show threadeyrea 🇨🇦2d ago

@harrymccracken And simply: writing is fun. Also aggravating sometimes, and challenging, but fun.

It's why the best personal use of robotics remains the robot vacuum cleaner.

Show threadJeff Atwood1d ago
@harrymccracken why is it all or nothing? Why not centaur it?
Show threadHarry McCracken1d ago
@codinghorror I use AI in all sorts of ways in my work, but the actual assembling of words I don’t want to turn over to someone (or something) else. Especially when my byline is on it.
Show threadJeff Atwood20h ago
@harrymccracken sure the centauring is just fine, would you turn over your whole column to another HUMAN? If so? Why???
Show threadHarry McCracken11h ago
@codinghorror As many of my editors would probably tell you, I am not always even that great with being edited. (I can read a published work of mine from ten years ago and spot the edits. Even when they’re good!)
Show threadjoseph1d ago
@harrymccracken @gruber wild that he just barely mentioned Wired's AI policy and didn't touch on the Benj Edwards story at Ars, which is owned by Wired's parent
Show threadRuby1d ago
@mixdup @harrymccracken @gruber 'he' or 'it', who knows?
Show threadMorgan Aldridge1d ago

@harrymccracken As a developer who also often enjoys writing, describing a bug/problem or a project scope is an important part of understanding & organizing my thoughts.

While writing and debugging actual code can be frustrating, time consuming, and exhausting, it's also the most interesting and **insightful** part of the process. You have to understand the project, the goal, and the previous developer's thought process (even if only your own.)

Similarly, I haven't written my best code yet!

Show threadHarry McCracken1d ago
@morgant I am loving vibecoding, but still am confronting the fact that I feel guilty I can’t code at that level on my own.
Show threadMorgan Aldridge16h ago
@harrymccracken I can certainly understand that and I think that's part of my underlying worry about the use of generative AI. Since you already think about the impact of using AI in other areas of your work and actively considering your thoughts/feelings when you do use AI, you're probably more likely to learn from the results of your vibecoding sessions. What percentage of others will do similar and will similar guilt dissuade them from improving themselves and instead rely on AI more?
Show threadMorgan Aldridge16h ago

@harrymccracken I have to admit that I haven't tried vibecoding yet, but partly because I've been turned off by results I've seen from the AI pull request reviews some of my clients have configured. AI does identify some quick stuff, but often misinterpret obvious things, and is wildly off on their actual business processes.

Of course, I also already battle with #ImposterSyndrome and that other developers are, "Better, faster, more productive," to lift a line from #Radiohead's #ParanoidAndroid.

Show threadMorgan Aldridge16h ago

@harrymccracken We're in this surreal stage of generative AI where we've made the prompt (and refinement) process a believable natural-language discussion with the AI. This makes it feel like we're interracting with a human, but the results that we're getting are superhuman. Hard not to compare oneself.

(While I say 'superhuman' I mean "super-average human." A friendly, agreeable, seemingly knowledgable person with super speed that gives shoddy results but will unceasingly refine the results.)

Show threadHarry McCracken11h ago
@morgant If you really know how to code, I can see how your reaction to vibecoding might be similar to my reaction to AI-written prose.
Show threadtoastedmilk22h ago
@harrymccracken the problem of “knowing what you want to say but not how to say it” often means you haven’t yet figured out what you want to say. if you turn to an LLM to help organize loose thoughts into prose, you’re outsourcing your cognition to a machine.
Show threadHarry McCracken11h ago
@Finin My solution has been to vibecode a word processor with outlining features built in, so I can begin to organize my thoughts before I begin writing in earnest. It helps!
Show threadtoastedmilk22h ago
@harrymccracken